diff --git a/.gitattributes b/.gitattributes index a6344aac8c09253b3b630fb776ae94478aa0275b..52373fe24473b1aa44333d318f578ae6bf04b49b 100644 --- a/.gitattributes +++ b/.gitattributes @@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text *.zip filter=lfs diff=lfs merge=lfs -text *.zst filter=lfs diff=lfs merge=lfs -text *tfevents* filter=lfs diff=lfs merge=lfs -text +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/config.json b/config.json new file mode 100644 index 0000000000000000000000000000000000000000..df5fabb1b3a6e42c28aa577ce27fb1662a5028a9 --- /dev/null +++ b/config.json @@ -0,0 +1,58 @@ +{ + "_name_or_path": "/data6/models/Meta-Llama-3.1-405B-Instruct", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "bos_token_id": 128000, + "eos_token_id": [ + 128001, + 128008, + 128009 + ], + "head_dim": 128, + "hidden_act": "silu", + "hidden_size": 16384, + "initializer_range": 0.02, + "intermediate_size": 53248, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "num_attention_heads": 128, + "num_hidden_layers": 126, + "num_key_value_heads": 16, + "pretraining_tp": 1, + "quantization_config": { + "amp": true, + "autoround_version": "0.4.3", + "backend": "auto_round:gptq:exllamav2", + "batch_size": 1, + "bits": 4, + "data_type": "int", + "dataset": "NeelNanda/pile-10k", + "enable_minmax_tuning": true, + "enable_norm_bias_tuning": false, + "enable_quanted_input": true, + "gradient_accumulate_steps": 4, + "group_size": -1, + "iters": 200, + "low_gpu_mem_usage": true, + "lr": 0.005, + "minmax_lr": 0.005, + "nsamples": 128, + "quant_method": "intel/auto-round", + "scale_dtype": "torch.float16", + "seqlen": 2048, + "sym": true, + "to_quant_block_names": null + }, + "rms_norm_eps": 1e-05, + "rope_scaling": null, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "torch_dtype": "bfloat16", + "transformers_version": "4.47.0", + "use_cache": true, + "vocab_size": 128256 +} diff --git a/generation_config.json b/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..0497cbb61179aa588915b4a56df71500e0abbc41 --- /dev/null +++ b/generation_config.json @@ -0,0 +1,12 @@ +{ + "bos_token_id": 128000, + "do_sample": true, + "eos_token_id": [ + 128001, + 128008, + 128009 + ], + "temperature": 0.6, + "top_p": 0.9, + "transformers_version": "4.47.0" +} diff --git a/model-00001-of-00044.safetensors b/model-00001-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dbe922a5b3f7eacd62dff3dcb444cf9a49a0db0d --- /dev/null +++ b/model-00001-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0e012ec3f44deea4607dbebbbe67e1ae7c0c16050cd756518548e30f7d65428 +size 4941445440 diff --git a/model-00002-of-00044.safetensors b/model-00002-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1f08c32898216b91a017cc225f7a63d991913d16 --- /dev/null +++ b/model-00002-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2f8259ff72a4e903925be56639246e4510ac4419cb11b6ff0cc25ad32b8746c +size 4835061656 diff --git a/model-00003-of-00044.safetensors b/model-00003-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b5bf4feab6bf6d2155c5eb056f8add908508539c --- /dev/null +++ b/model-00003-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:149f5e7eaadabc17813947121b978471dc7bc7338ffddf34fabda3e1111e6b98 +size 4835061656 diff --git a/model-00004-of-00044.safetensors b/model-00004-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2a6ac39be7a571baa26a65d66fdda5a08e91377b --- /dev/null +++ b/model-00004-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5167bf2c990e8f58a535eebd0c9c2ef2668e5115370f7757b71c18994de1e55d +size 4835061656 diff --git a/model-00005-of-00044.safetensors b/model-00005-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8ad59cc191833c0a17b8feeda1e11d20ef17bb5b --- /dev/null +++ b/model-00005-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d44db660ffa81ba4afbeab4ac5ca127499367b9a13d19878bbf58c9f18f4ee8 +size 4835061736 diff --git a/model-00006-of-00044.safetensors b/model-00006-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8197243e891809063292582c364468e90d7bfca7 --- /dev/null +++ b/model-00006-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c3855cf7aecdf9fae01a3632e9ee90a17434ac4de35f7227e6b4c5b78d03281 +size 4835061752 diff --git a/model-00007-of-00044.safetensors b/model-00007-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1aaa15d394a3b48058fc421ba267e016980728a6 --- /dev/null +++ b/model-00007-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d57299a546ab60d40a468751a330663338593ed74a9f5465be698ff2695bb0bb +size 4835061752 diff --git a/model-00008-of-00044.safetensors b/model-00008-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7c3b51a848a33eff81327dbfc6247c67445c90ca --- /dev/null +++ b/model-00008-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f47b6c03f6bcb2e59285a0a403b3504e88cccbe869d74c690f264869cf1a9d54 +size 4835061752 diff --git a/model-00009-of-00044.safetensors b/model-00009-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7d377a3d256cd93c87a48538e218647744700a71 --- /dev/null +++ b/model-00009-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e477431cc9ef00190fa9ef2954197fb39068ea162aa3d160d695d65a9aa93dac +size 4835061752 diff --git a/model-00010-of-00044.safetensors b/model-00010-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..06c8508911bb6a92109b47f4d02993aeb28e1071 --- /dev/null +++ b/model-00010-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9d1ddf3a47512dac382a0b3546950c2e5863ee228c8945cb5e2bee5d578f394 +size 4835061752 diff --git a/model-00011-of-00044.safetensors b/model-00011-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9efa43cfbcde386efa0b78d15cb02c887e2a352d --- /dev/null +++ b/model-00011-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0c2e19f2bfb9bfe837f2e8ee4d7819e202e2db3030a5d9feeec8b8cf0b15e48 +size 4835061752 diff --git a/model-00012-of-00044.safetensors b/model-00012-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f779d23f9f30ae5330a6bf5c2d371a369a5dd609 --- /dev/null +++ b/model-00012-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c11ec4493cf163310f26029858759d236d04b0f6378ec4eb7cbf6283821c8e8 +size 4835061752 diff --git a/model-00013-of-00044.safetensors b/model-00013-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c4ef5e9f527eb49c53e6abdcff254d3d5e1b573f --- /dev/null +++ b/model-00013-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a206bf08b0b9048548879e3046e07361e22ff992df307488a540d2df1f4f82f8 +size 4835061752 diff --git a/model-00014-of-00044.safetensors b/model-00014-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9ce5ed5b7c896dafa62e012d41133fbe4ce64601 --- /dev/null +++ b/model-00014-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0252d569d041e4c5888f71b756c9c730f7c3fadfef597f51587aa1b9099d46f8 +size 4835061752 diff --git a/model-00015-of-00044.safetensors b/model-00015-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3b9af8328d38e2969c96c7efc30c153551ad7263 --- /dev/null +++ b/model-00015-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:debb9846b20a6cda398269af579dd3da5f293143507c804ed02d57a4d3c2e867 +size 4835061752 diff --git a/model-00016-of-00044.safetensors b/model-00016-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3cd3246de48ab3c308a74e85cb4a4d27e8e94232 --- /dev/null +++ b/model-00016-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:356ffcfa18ac44c18f5557391279d869bc5986deca193f3e7f385ddbc0e40f48 +size 4835061752 diff --git a/model-00017-of-00044.safetensors b/model-00017-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4b434193711b2a5e62ec0dafac5f3685b2f25001 --- /dev/null +++ b/model-00017-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41e3b385c7345b5d820f28b6114ba9e212632ac7a785572ae6933515717f96cc +size 4835061752 diff --git a/model-00018-of-00044.safetensors b/model-00018-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f9a524162c3bfde2aaa4d9452389eb74de68ddee --- /dev/null +++ b/model-00018-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a80d005a53159a2fe5fcd1d65d0b1e3119cd0a50ef0392868019f5303fc92fb +size 4835061752 diff --git a/model-00019-of-00044.safetensors b/model-00019-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6c1f794080288c7880a3d88a90f116402534a4ed --- /dev/null +++ b/model-00019-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5edd68e3305fa4d6c535efb69369093e42d4841ba61ff202f515791426fa57f0 +size 4835061752 diff --git a/model-00020-of-00044.safetensors b/model-00020-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..81d2abbca0cb6c62649c1053e70a5446d8c10648 --- /dev/null +++ b/model-00020-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49e1956c7d365141cedda43d5f62bffd60b6812c22fd4912746cafa9bce0d357 +size 4835061752 diff --git a/model-00021-of-00044.safetensors b/model-00021-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c91adc66e72f73dae68467d12151f06522a268dd --- /dev/null +++ b/model-00021-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27edf55ba8497284bb2684a267efa8adfdc10af2f2cdebd46bdb73f32727b6be +size 4835061752 diff --git a/model-00022-of-00044.safetensors b/model-00022-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..85e0c8875ffa212f71518375425f17d2abdd767b --- /dev/null +++ b/model-00022-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e52d195e2b475c3b632b66e3311ad056d0b6b58853d37df94f826e25ead431bd +size 4835061752 diff --git a/model-00023-of-00044.safetensors b/model-00023-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a17dcea0bf7b949a3038eb28da9f19cf018fb3c0 --- /dev/null +++ b/model-00023-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db951f808f3944001357bf8e5bf4aa6fcb022a9d1685966872b729d72e744b8d +size 4835061752 diff --git a/model-00024-of-00044.safetensors b/model-00024-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6c2494b4bf916241440a8e03d6d8c676e801e7f2 --- /dev/null +++ b/model-00024-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae76848da89b0300f8e649d1c49c9de351723c2692c80bc53715567e6165a92b +size 4835061752 diff --git a/model-00025-of-00044.safetensors b/model-00025-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f8dbfa0eae2930d79654556c74e0ea3f257d9a8b --- /dev/null +++ b/model-00025-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09417331a6dacd03dc6273a5e62917659dd9c7b030e8f90d88354619cb4d6257 +size 4835061752 diff --git a/model-00026-of-00044.safetensors b/model-00026-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bcd6d2cedc6afa08790edb51ec4dce2ac432e9ed --- /dev/null +++ b/model-00026-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef43240ab186dd84029eb2e2c78843be4c0c9b4acb85ed875821e62d98d2630c +size 4835061752 diff --git a/model-00027-of-00044.safetensors b/model-00027-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ca12f3fe817b33c1721d47583b5eb7d9fbe8bea0 --- /dev/null +++ b/model-00027-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9dc748768c9df33ae5028f9cb9382bdc938c058b47a9001da5cd73309b4ac348 +size 4835061752 diff --git a/model-00028-of-00044.safetensors b/model-00028-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..74e3e3db18bf228c77e4ecb0d3a7cc762acd9849 --- /dev/null +++ b/model-00028-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd52c89e204eb68f397ecab0264591d3394e972c31b4a4b7b332a71b637b1713 +size 4835061752 diff --git a/model-00029-of-00044.safetensors b/model-00029-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9f8f6057905c2418ed687ef6aac2ce7eba836e6f --- /dev/null +++ b/model-00029-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d98a912a986079821863e330ba02b0230a2ca4ebd3594aedf5ada74685a8a339 +size 4835061752 diff --git a/model-00030-of-00044.safetensors b/model-00030-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d8f779269d5bb7665441c0b6083c7ac73ae97713 --- /dev/null +++ b/model-00030-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e27bd3d17d644d72c8a93ae3a7f544f56b72b13b075d0ceb1283a439b13d8397 +size 4835061752 diff --git a/model-00031-of-00044.safetensors b/model-00031-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..425fdadbf728ee7eb28b44291de212b1be1b8ae9 --- /dev/null +++ b/model-00031-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7dd207b62db6f459c3142cacd4316078f7cfed927a5c1b5dd586598bde4e66c +size 4835061752 diff --git a/model-00032-of-00044.safetensors b/model-00032-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fbc52bd9830edfac41f5ae5a6768c1691574fcb5 --- /dev/null +++ b/model-00032-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:363f75e93850dd5bbf21085c46998d2de4be8fe7ea3e6df86be2bb01c60320a3 +size 4835061752 diff --git a/model-00033-of-00044.safetensors b/model-00033-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1b6245d0d3b5a3faac40d4c2912ea0473543cf23 --- /dev/null +++ b/model-00033-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0a6f8859542abce3a64b8490dcce0a74577f7e68a792dc9ee9d2d977a2e81ec +size 4835061752 diff --git a/model-00034-of-00044.safetensors b/model-00034-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cfeb382fc8c35d6d3b94f1bc712aee15d19fa42c --- /dev/null +++ b/model-00034-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d051d247352a1176b33fdede0e1fb8d20863cd5bc6e6b3ca9b1c7d2f1e3f475d +size 4835061752 diff --git a/model-00035-of-00044.safetensors b/model-00035-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f4705ddee7144b73f8ddf2e4a8de2cea71e93f93 --- /dev/null +++ b/model-00035-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d27196eb6805942f7fe28e98ed367e836e0bb9218923e7538ea4e09fb7fd31c1 +size 4835061832 diff --git a/model-00036-of-00044.safetensors b/model-00036-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..842be7efafae1da2058b011e8d3ecd4258a812ff --- /dev/null +++ b/model-00036-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87f2fea2a9d710060eb986ea952f5feb1ce948b229236ca499687146df196032 +size 4835061840 diff --git a/model-00037-of-00044.safetensors b/model-00037-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e1b7be10fb74f41988074d62977d0eb546358d7b --- /dev/null +++ b/model-00037-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5db7c3d7fa72c39de61c7c1aced4528eb7cafb5fbc32546e63ee7b8864f8d192 +size 4835061840 diff --git a/model-00038-of-00044.safetensors b/model-00038-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e14963561d340c33a4a3e178ac459df3162cbdf3 --- /dev/null +++ b/model-00038-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88ad6d6599bc4299ca5a292c3863fc8a6cdc86ffe5d151fb914494ed685a9ae3 +size 4835061840 diff --git a/model-00039-of-00044.safetensors b/model-00039-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3df27705c80e3afa601e8ded5763516e82d301b5 --- /dev/null +++ b/model-00039-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:79fba7636576365e5b266a55cd1c56b2d40c18f1fb98dc52abb52a1c136ca6a4 +size 4835061840 diff --git a/model-00040-of-00044.safetensors b/model-00040-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..07fa11a8172bc924c1878bb6a6d0d5e104cfaeb0 --- /dev/null +++ b/model-00040-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d570b4ccaf60a7f4d0cfbb999dae5f10fb47b6e35083804469a764bada4dee66 +size 4835061840 diff --git a/model-00041-of-00044.safetensors b/model-00041-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4e40a009056ff06cc1d1efabf80eb86b18585566 --- /dev/null +++ b/model-00041-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8660c30ebc4ce89e1d9f71fd5049ff8b9001d0efc9fc9ec0a7c5b51de1c244f +size 4835061840 diff --git a/model-00042-of-00044.safetensors b/model-00042-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ddcca6cd8997ff5cb1320cd3ba0155ded3c1a1b4 --- /dev/null +++ b/model-00042-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:475c7145b02f53c208e53add8e3defcdfd08d6283d71d498ef21a3a52c7d11b5 +size 4835061840 diff --git a/model-00043-of-00044.safetensors b/model-00043-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4e7d3528d7aeb51838281383cb65690fa20062ae --- /dev/null +++ b/model-00043-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:051d02b61978157fb7820ce51a20b6918d56835a6cc856e72dfac5cf675dfc35 +size 4096341936 diff --git a/model-00044-of-00044.safetensors b/model-00044-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..189b2ba970360412785980a181e8aa7ae7624a18 --- /dev/null +++ b/model-00044-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71b079a02069ebe5a5f4f8135815f136afb1d6fc4423620dcb2a14fa399cdf12 +size 4202692736 diff --git a/model.safetensors.index.json b/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..071ec3771d778f55c9755e4d153966fa4c49105c --- /dev/null +++ b/model.safetensors.index.json @@ -0,0 +1,3790 @@ +{ + "metadata": { + "total_size": 211477581824 + }, + "weight_map": { + "lm_head.weight": "model-00044-of-00044.safetensors", + "model.embed_tokens.weight": "model-00001-of-00044.safetensors", + "model.layers.0.input_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.g_idx": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.qweight": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.qzeros": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.scales": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.up_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.up_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.up_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.up_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.0.post_attention_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.g_idx": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.qweight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.qzeros": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.scales": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.g_idx": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.qweight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.qzeros": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.scales": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.g_idx": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.qweight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.qzeros": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.scales": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.g_idx": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.qweight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.qzeros": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.scales": "model-00001-of-00044.safetensors", + "model.layers.1.input_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.1.post_attention_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.10.input_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.10.post_attention_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.100.input_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.100.post_attention_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.input_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.post_attention_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.102.input_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.up_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.up_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.up_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.up_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.102.post_attention_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.103.input_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.103.post_attention_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.input_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.post_attention_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.105.input_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.up_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.up_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.up_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.up_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.105.post_attention_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.g_idx": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.qweight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.qzeros": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.scales": "model-00036-of-00044.safetensors", + "model.layers.106.input_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.106.post_attention_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.input_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.post_attention_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.108.input_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.up_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.up_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.up_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.up_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.108.post_attention_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.g_idx": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.qweight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.qzeros": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.scales": "model-00037-of-00044.safetensors", + "model.layers.109.input_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.109.post_attention_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.11.input_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.11.post_attention_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.110.input_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.110.post_attention_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.111.input_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.up_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.up_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.up_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.up_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.111.post_attention_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.g_idx": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.qweight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.qzeros": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.scales": "model-00038-of-00044.safetensors", + "model.layers.112.input_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.112.post_attention_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.input_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.post_attention_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.114.input_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.up_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.up_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.up_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.up_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.114.post_attention_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.g_idx": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.qweight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.qzeros": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.scales": "model-00039-of-00044.safetensors", + "model.layers.115.input_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.115.post_attention_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.input_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.post_attention_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.117.input_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.up_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.up_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.up_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.up_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.117.post_attention_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.g_idx": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.qweight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.qzeros": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.scales": "model-00040-of-00044.safetensors", + "model.layers.118.input_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.118.post_attention_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.input_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.post_attention_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.12.input_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.up_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.up_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.up_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.up_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.12.post_attention_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.120.input_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.up_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.up_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.up_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.up_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.120.post_attention_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.g_idx": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.qweight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.qzeros": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.scales": "model-00041-of-00044.safetensors", + "model.layers.121.input_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.121.post_attention_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.input_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.post_attention_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.123.input_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.up_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.up_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.up_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.up_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.123.post_attention_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.g_idx": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.qweight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.qzeros": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.scales": "model-00042-of-00044.safetensors", + "model.layers.124.input_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.124.post_attention_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.input_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.post_attention_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.g_idx": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.qweight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.qzeros": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.scales": "model-00043-of-00044.safetensors", + "model.layers.13.input_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.13.post_attention_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.input_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.post_attention_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.15.input_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.up_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.up_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.up_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.up_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.15.post_attention_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.g_idx": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.qweight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.qzeros": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.scales": "model-00006-of-00044.safetensors", + "model.layers.16.input_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.16.post_attention_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.input_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.post_attention_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.18.input_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.up_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.up_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.up_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.up_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.18.post_attention_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.g_idx": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.qweight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.qzeros": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.scales": "model-00007-of-00044.safetensors", + "model.layers.19.input_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.19.post_attention_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.2.input_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.2.post_attention_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.20.input_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.20.post_attention_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.21.input_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.up_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.up_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.up_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.up_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.21.post_attention_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.g_idx": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.qweight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.qzeros": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.scales": "model-00008-of-00044.safetensors", + "model.layers.22.input_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.22.post_attention_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.input_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.post_attention_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.24.input_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.up_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.up_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.up_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.up_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.24.post_attention_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.g_idx": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.qweight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.qzeros": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.scales": "model-00009-of-00044.safetensors", + "model.layers.25.input_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.25.post_attention_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.input_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.post_attention_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.27.input_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.up_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.up_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.up_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.up_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.27.post_attention_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.g_idx": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.qweight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.qzeros": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.scales": "model-00010-of-00044.safetensors", + "model.layers.28.input_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.28.post_attention_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.input_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.post_attention_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.3.input_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.up_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.up_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.up_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.up_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.3.post_attention_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.g_idx": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.qweight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.qzeros": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.scales": "model-00002-of-00044.safetensors", + "model.layers.30.input_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.up_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.up_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.up_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.up_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.30.post_attention_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.g_idx": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.qweight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.qzeros": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.scales": "model-00011-of-00044.safetensors", + "model.layers.31.input_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.31.post_attention_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.input_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.post_attention_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.33.input_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.up_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.up_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.up_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.up_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.33.post_attention_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.g_idx": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.qweight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.qzeros": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.scales": "model-00012-of-00044.safetensors", + "model.layers.34.input_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.34.post_attention_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.input_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.post_attention_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.36.input_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.up_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.up_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.up_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.up_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.36.post_attention_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.g_idx": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.qweight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.qzeros": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.scales": "model-00013-of-00044.safetensors", + "model.layers.37.input_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.37.post_attention_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.input_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.post_attention_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.39.input_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.up_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.up_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.up_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.up_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.39.post_attention_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.g_idx": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.qweight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.qzeros": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.scales": "model-00014-of-00044.safetensors", + "model.layers.4.input_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.4.post_attention_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.40.input_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.40.post_attention_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.input_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.post_attention_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.42.input_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.up_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.up_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.up_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.up_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.42.post_attention_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.g_idx": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.qweight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.qzeros": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.scales": "model-00015-of-00044.safetensors", + "model.layers.43.input_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.43.post_attention_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.input_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.post_attention_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.45.input_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.up_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.up_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.up_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.up_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.45.post_attention_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.g_idx": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.qweight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.qzeros": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.scales": "model-00016-of-00044.safetensors", + "model.layers.46.input_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.46.post_attention_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.input_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.post_attention_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.48.input_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.up_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.up_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.up_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.up_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.48.post_attention_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.g_idx": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.qweight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.qzeros": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.scales": "model-00017-of-00044.safetensors", + "model.layers.49.input_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.49.post_attention_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.5.input_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.5.post_attention_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.50.input_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.50.post_attention_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.51.input_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.up_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.up_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.up_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.up_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.51.post_attention_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.g_idx": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.qweight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.qzeros": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.scales": "model-00018-of-00044.safetensors", + "model.layers.52.input_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.52.post_attention_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.input_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.post_attention_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.54.input_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.up_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.up_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.up_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.up_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.54.post_attention_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.g_idx": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.qweight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.qzeros": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.scales": "model-00019-of-00044.safetensors", + "model.layers.55.input_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.55.post_attention_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.input_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.post_attention_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.57.input_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.up_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.up_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.up_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.up_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.57.post_attention_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.g_idx": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.qweight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.qzeros": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.scales": "model-00020-of-00044.safetensors", + "model.layers.58.input_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.58.post_attention_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.input_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.post_attention_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.6.input_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.up_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.up_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.up_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.up_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.6.post_attention_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.g_idx": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.qweight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.qzeros": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.scales": "model-00003-of-00044.safetensors", + "model.layers.60.input_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.up_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.up_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.up_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.up_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.60.post_attention_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.g_idx": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.qweight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.qzeros": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.scales": "model-00021-of-00044.safetensors", + "model.layers.61.input_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.61.post_attention_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.input_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.post_attention_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.63.input_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.up_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.up_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.up_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.up_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.63.post_attention_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.g_idx": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.qweight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.qzeros": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.scales": "model-00022-of-00044.safetensors", + "model.layers.64.input_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.64.post_attention_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.input_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.post_attention_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.66.input_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.up_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.up_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.up_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.up_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.66.post_attention_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.g_idx": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.qweight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.qzeros": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.scales": "model-00023-of-00044.safetensors", + "model.layers.67.input_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.67.post_attention_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.input_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.post_attention_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.69.input_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.up_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.up_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.up_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.up_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.69.post_attention_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.g_idx": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.qweight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.qzeros": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.scales": "model-00024-of-00044.safetensors", + "model.layers.7.input_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.7.post_attention_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.70.input_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.70.post_attention_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.input_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.post_attention_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.72.input_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.up_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.up_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.up_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.up_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.72.post_attention_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.g_idx": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.qweight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.qzeros": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.scales": "model-00025-of-00044.safetensors", + "model.layers.73.input_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.73.post_attention_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.input_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.post_attention_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.75.input_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.up_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.up_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.up_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.up_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.75.post_attention_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.g_idx": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.qweight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.qzeros": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.scales": "model-00026-of-00044.safetensors", + "model.layers.76.input_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.76.post_attention_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.input_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.post_attention_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.78.input_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.up_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.up_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.up_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.up_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.78.post_attention_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.g_idx": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.qweight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.qzeros": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.scales": "model-00027-of-00044.safetensors", + "model.layers.79.input_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.79.post_attention_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.8.input_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.8.post_attention_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.80.input_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.80.post_attention_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.81.input_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.up_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.up_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.up_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.up_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.81.post_attention_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.g_idx": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.qweight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.qzeros": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.scales": "model-00028-of-00044.safetensors", + "model.layers.82.input_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.82.post_attention_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.input_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.post_attention_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.84.input_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.up_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.up_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.up_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.up_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.84.post_attention_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.g_idx": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.qweight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.qzeros": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.scales": "model-00029-of-00044.safetensors", + "model.layers.85.input_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.85.post_attention_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.input_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.post_attention_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.87.input_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.up_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.up_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.up_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.up_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.87.post_attention_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.g_idx": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.qweight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.qzeros": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.scales": "model-00030-of-00044.safetensors", + "model.layers.88.input_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.88.post_attention_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.input_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.post_attention_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.9.input_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.up_proj.g_idx": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.up_proj.qweight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.up_proj.qzeros": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.up_proj.scales": "model-00005-of-00044.safetensors", + "model.layers.9.post_attention_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.g_idx": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.qweight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.qzeros": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.scales": "model-00004-of-00044.safetensors", + "model.layers.90.input_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.up_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.up_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.up_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.up_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.90.post_attention_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.g_idx": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.qweight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.qzeros": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.scales": "model-00031-of-00044.safetensors", + "model.layers.91.input_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.91.post_attention_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.input_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.post_attention_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.93.input_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.up_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.up_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.up_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.up_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.93.post_attention_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.g_idx": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.qweight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.qzeros": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.scales": "model-00032-of-00044.safetensors", + "model.layers.94.input_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.94.post_attention_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.input_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.post_attention_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.96.input_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.up_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.up_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.up_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.up_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.96.post_attention_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.g_idx": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.qweight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.qzeros": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.scales": "model-00033-of-00044.safetensors", + "model.layers.97.input_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.97.post_attention_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.input_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.post_attention_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.99.input_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.up_proj.g_idx": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.up_proj.qweight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.up_proj.qzeros": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.up_proj.scales": "model-00035-of-00044.safetensors", + "model.layers.99.post_attention_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.scales": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.g_idx": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.qweight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.qzeros": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.scales": "model-00034-of-00044.safetensors", + "model.norm.weight": "model-00043-of-00044.safetensors" + } +} diff --git a/quantization_config.json b/quantization_config.json new file mode 100644 index 0000000000000000000000000000000000000000..c980b2f3360e27b0073ad14ad707bb2c169a8d4a --- /dev/null +++ b/quantization_config.json @@ -0,0 +1,24 @@ +{ + "bits": 4, + "group_size": -1, + "sym": true, + "data_type": "int", + "enable_quanted_input": true, + "enable_minmax_tuning": true, + "seqlen": 2048, + "batch_size": 1, + "scale_dtype": "torch.float16", + "lr": 0.005, + "minmax_lr": 0.005, + "gradient_accumulate_steps": 4, + "iters": 200, + "amp": true, + "nsamples": 128, + "low_gpu_mem_usage": true, + "to_quant_block_names": null, + "enable_norm_bias_tuning": false, + "dataset": "NeelNanda/pile-10k", + "autoround_version": "0.4.3", + "quant_method": "intel/auto-round", + "backend": "auto_round:gptq:exllamav2" +} \ No newline at end of file diff --git a/special_tokens_map.json b/special_tokens_map.json new file mode 100644 index 0000000000000000000000000000000000000000..02ee80b6196926a5ad790a004d9efd6ab1ba6542 --- /dev/null +++ b/special_tokens_map.json @@ -0,0 +1,16 @@ +{ + "bos_token": { + "content": "<|begin_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false + }, + "eos_token": { + "content": "<|eot_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false + } +} diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..af2d22fff23798bea98b1730ae7cdacaee0b087a --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ade1dac458f86f9bea8bf35b713f14e1bbed24228429534038e9f7e54ea3e8b6 +size 17208712 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..41c49944488353226130158c621aa15b1eaa7788 --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,2063 @@ +{ + "added_tokens_decoder": { + "128000": { + "content": "<|begin_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128001": { + "content": "<|end_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128002": { + "content": "<|reserved_special_token_0|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128003": { + "content": "<|reserved_special_token_1|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128004": { + "content": "<|finetune_right_pad_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128005": { + "content": "<|reserved_special_token_2|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128006": { + "content": "<|start_header_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128007": { + "content": "<|end_header_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128008": { + "content": "<|eom_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128009": { + "content": "<|eot_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128010": { + "content": "<|python_tag|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128011": { + "content": "<|reserved_special_token_3|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128012": { + "content": "<|reserved_special_token_4|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128013": { + "content": "<|reserved_special_token_5|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128014": { + "content": "<|reserved_special_token_6|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128015": { + "content": "<|reserved_special_token_7|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128016": { + "content": "<|reserved_special_token_8|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128017": { + "content": "<|reserved_special_token_9|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128018": { + "content": "<|reserved_special_token_10|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128019": { + "content": "<|reserved_special_token_11|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128020": { + "content": "<|reserved_special_token_12|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128021": { + "content": "<|reserved_special_token_13|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128022": { + "content": "<|reserved_special_token_14|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128023": { + "content": "<|reserved_special_token_15|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128024": { + "content": "<|reserved_special_token_16|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128025": { + "content": "<|reserved_special_token_17|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128026": { + "content": "<|reserved_special_token_18|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128027": { + "content": "<|reserved_special_token_19|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128028": { + "content": "<|reserved_special_token_20|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128029": { + "content": "<|reserved_special_token_21|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128030": { + "content": "<|reserved_special_token_22|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128031": { + "content": "<|reserved_special_token_23|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128032": { + "content": "<|reserved_special_token_24|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128033": { + "content": "<|reserved_special_token_25|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128034": { + "content": "<|reserved_special_token_26|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128035": { + "content": "<|reserved_special_token_27|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128036": { + "content": "<|reserved_special_token_28|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128037": { + "content": "<|reserved_special_token_29|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128038": { + "content": "<|reserved_special_token_30|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128039": { + "content": "<|reserved_special_token_31|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128040": { + "content": "<|reserved_special_token_32|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128041": { + "content": "<|reserved_special_token_33|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128042": { + "content": "<|reserved_special_token_34|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128043": { + "content": "<|reserved_special_token_35|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128044": { + "content": "<|reserved_special_token_36|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128045": { + "content": "<|reserved_special_token_37|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128046": { + "content": "<|reserved_special_token_38|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128047": { + "content": "<|reserved_special_token_39|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128048": { + "content": "<|reserved_special_token_40|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128049": { + "content": "<|reserved_special_token_41|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128050": { + "content": "<|reserved_special_token_42|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128051": { + "content": "<|reserved_special_token_43|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128052": { + "content": "<|reserved_special_token_44|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128053": { + "content": "<|reserved_special_token_45|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128054": { + "content": "<|reserved_special_token_46|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128055": { + "content": "<|reserved_special_token_47|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128056": { + "content": "<|reserved_special_token_48|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128057": { + "content": "<|reserved_special_token_49|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128058": { + "content": "<|reserved_special_token_50|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128059": { + "content": "<|reserved_special_token_51|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128060": { + "content": "<|reserved_special_token_52|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128061": { + "content": "<|reserved_special_token_53|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128062": { + "content": "<|reserved_special_token_54|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128063": { + "content": "<|reserved_special_token_55|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128064": { + "content": "<|reserved_special_token_56|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128065": { + "content": "<|reserved_special_token_57|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128066": { + "content": "<|reserved_special_token_58|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128067": { + "content": "<|reserved_special_token_59|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128068": { + "content": "<|reserved_special_token_60|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128069": { + "content": "<|reserved_special_token_61|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128070": { + "content": "<|reserved_special_token_62|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128071": { + "content": "<|reserved_special_token_63|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128072": { + "content": "<|reserved_special_token_64|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128073": { + "content": "<|reserved_special_token_65|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128074": { + "content": "<|reserved_special_token_66|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128075": { + "content": "<|reserved_special_token_67|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128076": { + "content": "<|reserved_special_token_68|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128077": { + "content": "<|reserved_special_token_69|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128078": { + "content": "<|reserved_special_token_70|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128079": { + "content": "<|reserved_special_token_71|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128080": { + "content": "<|reserved_special_token_72|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128081": { + "content": "<|reserved_special_token_73|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128082": { + "content": "<|reserved_special_token_74|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128083": { + "content": "<|reserved_special_token_75|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128084": { + "content": "<|reserved_special_token_76|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128085": { + "content": "<|reserved_special_token_77|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128086": { + "content": "<|reserved_special_token_78|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128087": { + "content": "<|reserved_special_token_79|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128088": { + "content": "<|reserved_special_token_80|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128089": { + "content": "<|reserved_special_token_81|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128090": { + "content": "<|reserved_special_token_82|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128091": { + "content": "<|reserved_special_token_83|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128092": { + "content": "<|reserved_special_token_84|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128093": { + "content": "<|reserved_special_token_85|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128094": { + "content": "<|reserved_special_token_86|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128095": { + "content": "<|reserved_special_token_87|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128096": { + "content": "<|reserved_special_token_88|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128097": { + "content": "<|reserved_special_token_89|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128098": { + "content": "<|reserved_special_token_90|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128099": { + "content": "<|reserved_special_token_91|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128100": { + "content": "<|reserved_special_token_92|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128101": { + "content": "<|reserved_special_token_93|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128102": { + "content": "<|reserved_special_token_94|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128103": { + "content": "<|reserved_special_token_95|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128104": { + "content": "<|reserved_special_token_96|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128105": { + "content": "<|reserved_special_token_97|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128106": { + "content": "<|reserved_special_token_98|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128107": { + "content": "<|reserved_special_token_99|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128108": { + "content": "<|reserved_special_token_100|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128109": { + "content": "<|reserved_special_token_101|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128110": { + "content": "<|reserved_special_token_102|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128111": { + "content": "<|reserved_special_token_103|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128112": { + "content": "<|reserved_special_token_104|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128113": { + "content": "<|reserved_special_token_105|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128114": { + "content": "<|reserved_special_token_106|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128115": { + "content": "<|reserved_special_token_107|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128116": { + "content": "<|reserved_special_token_108|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128117": { + "content": "<|reserved_special_token_109|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128118": { + "content": "<|reserved_special_token_110|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128119": { + "content": "<|reserved_special_token_111|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128120": { + "content": "<|reserved_special_token_112|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128121": { + "content": "<|reserved_special_token_113|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128122": { + "content": "<|reserved_special_token_114|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128123": { + "content": "<|reserved_special_token_115|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128124": { + "content": "<|reserved_special_token_116|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128125": { + "content": "<|reserved_special_token_117|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128126": { + "content": "<|reserved_special_token_118|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128127": { + "content": "<|reserved_special_token_119|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128128": { + "content": "<|reserved_special_token_120|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128129": { + "content": "<|reserved_special_token_121|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128130": { + "content": "<|reserved_special_token_122|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128131": { + "content": "<|reserved_special_token_123|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128132": { + "content": "<|reserved_special_token_124|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128133": { + "content": "<|reserved_special_token_125|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128134": { + "content": "<|reserved_special_token_126|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128135": { + "content": "<|reserved_special_token_127|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128136": { + "content": "<|reserved_special_token_128|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128137": { + "content": "<|reserved_special_token_129|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128138": { + "content": "<|reserved_special_token_130|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128139": { + "content": "<|reserved_special_token_131|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128140": { + "content": "<|reserved_special_token_132|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128141": { + "content": "<|reserved_special_token_133|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128142": { + "content": "<|reserved_special_token_134|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128143": { + "content": "<|reserved_special_token_135|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128144": { + "content": "<|reserved_special_token_136|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128145": { + "content": "<|reserved_special_token_137|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128146": { + "content": "<|reserved_special_token_138|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128147": { + "content": "<|reserved_special_token_139|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128148": { + "content": "<|reserved_special_token_140|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128149": { + "content": "<|reserved_special_token_141|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128150": { + "content": "<|reserved_special_token_142|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128151": { + "content": "<|reserved_special_token_143|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128152": { + "content": "<|reserved_special_token_144|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128153": { + "content": "<|reserved_special_token_145|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128154": { + "content": "<|reserved_special_token_146|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128155": { + "content": "<|reserved_special_token_147|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128156": { + "content": "<|reserved_special_token_148|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128157": { + "content": "<|reserved_special_token_149|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128158": { + "content": "<|reserved_special_token_150|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128159": { + "content": "<|reserved_special_token_151|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128160": { + "content": "<|reserved_special_token_152|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128161": { + "content": "<|reserved_special_token_153|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128162": { + "content": "<|reserved_special_token_154|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128163": { + "content": "<|reserved_special_token_155|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128164": { + "content": "<|reserved_special_token_156|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128165": { + "content": "<|reserved_special_token_157|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128166": { + "content": "<|reserved_special_token_158|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128167": { + "content": "<|reserved_special_token_159|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128168": { + "content": "<|reserved_special_token_160|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128169": { + "content": "<|reserved_special_token_161|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128170": { + "content": "<|reserved_special_token_162|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128171": { + "content": "<|reserved_special_token_163|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128172": { + "content": "<|reserved_special_token_164|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128173": { + "content": "<|reserved_special_token_165|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128174": { + "content": "<|reserved_special_token_166|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128175": { + "content": "<|reserved_special_token_167|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128176": { + "content": "<|reserved_special_token_168|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128177": { + "content": "<|reserved_special_token_169|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128178": { + "content": "<|reserved_special_token_170|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128179": { + "content": "<|reserved_special_token_171|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128180": { + "content": "<|reserved_special_token_172|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128181": { + "content": "<|reserved_special_token_173|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128182": { + "content": "<|reserved_special_token_174|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128183": { + "content": "<|reserved_special_token_175|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128184": { + "content": "<|reserved_special_token_176|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128185": { + "content": "<|reserved_special_token_177|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128186": { + "content": "<|reserved_special_token_178|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128187": { + "content": "<|reserved_special_token_179|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128188": { + "content": "<|reserved_special_token_180|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128189": { + "content": "<|reserved_special_token_181|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128190": { + "content": "<|reserved_special_token_182|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128191": { + "content": "<|reserved_special_token_183|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128192": { + "content": "<|reserved_special_token_184|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128193": { + "content": "<|reserved_special_token_185|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128194": { + "content": "<|reserved_special_token_186|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128195": { + "content": "<|reserved_special_token_187|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128196": { + "content": "<|reserved_special_token_188|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128197": { + "content": "<|reserved_special_token_189|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128198": { + "content": "<|reserved_special_token_190|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128199": { + "content": "<|reserved_special_token_191|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128200": { + "content": "<|reserved_special_token_192|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128201": { + "content": "<|reserved_special_token_193|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128202": { + "content": "<|reserved_special_token_194|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128203": { + "content": "<|reserved_special_token_195|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128204": { + "content": "<|reserved_special_token_196|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128205": { + "content": "<|reserved_special_token_197|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128206": { + "content": "<|reserved_special_token_198|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128207": { + "content": "<|reserved_special_token_199|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128208": { + "content": "<|reserved_special_token_200|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128209": { + "content": "<|reserved_special_token_201|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128210": { + "content": "<|reserved_special_token_202|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128211": { + "content": "<|reserved_special_token_203|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128212": { + "content": "<|reserved_special_token_204|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128213": { + "content": "<|reserved_special_token_205|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128214": { + "content": "<|reserved_special_token_206|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128215": { + "content": "<|reserved_special_token_207|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128216": { + "content": "<|reserved_special_token_208|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128217": { + "content": "<|reserved_special_token_209|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128218": { + "content": "<|reserved_special_token_210|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128219": { + "content": "<|reserved_special_token_211|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128220": { + "content": "<|reserved_special_token_212|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128221": { + "content": "<|reserved_special_token_213|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128222": { + "content": "<|reserved_special_token_214|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128223": { + "content": "<|reserved_special_token_215|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128224": { + "content": "<|reserved_special_token_216|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128225": { + "content": "<|reserved_special_token_217|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128226": { + "content": "<|reserved_special_token_218|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128227": { + "content": "<|reserved_special_token_219|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128228": { + "content": "<|reserved_special_token_220|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128229": { + "content": "<|reserved_special_token_221|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128230": { + "content": "<|reserved_special_token_222|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128231": { + "content": "<|reserved_special_token_223|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128232": { + "content": "<|reserved_special_token_224|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128233": { + "content": "<|reserved_special_token_225|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128234": { + "content": "<|reserved_special_token_226|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128235": { + "content": "<|reserved_special_token_227|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128236": { + "content": "<|reserved_special_token_228|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128237": { + "content": "<|reserved_special_token_229|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128238": { + "content": "<|reserved_special_token_230|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128239": { + "content": "<|reserved_special_token_231|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128240": { + "content": "<|reserved_special_token_232|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128241": { + "content": "<|reserved_special_token_233|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128242": { + "content": "<|reserved_special_token_234|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128243": { + "content": "<|reserved_special_token_235|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128244": { + "content": "<|reserved_special_token_236|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128245": { + "content": "<|reserved_special_token_237|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128246": { + "content": "<|reserved_special_token_238|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128247": { + "content": "<|reserved_special_token_239|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128248": { + "content": "<|reserved_special_token_240|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128249": { + "content": "<|reserved_special_token_241|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128250": { + "content": "<|reserved_special_token_242|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128251": { + "content": "<|reserved_special_token_243|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128252": { + "content": "<|reserved_special_token_244|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128253": { + "content": "<|reserved_special_token_245|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128254": { + "content": "<|reserved_special_token_246|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128255": { + "content": "<|reserved_special_token_247|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + } + }, + "bos_token": "<|begin_of_text|>", + "chat_template": "{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}", + "clean_up_tokenization_spaces": true, + "eos_token": "<|eot_id|>", + "extra_special_tokens": {}, + "model_input_names": [ + "input_ids", + "attention_mask" + ], + "model_max_length": 131072, + "tokenizer_class": "PreTrainedTokenizerFast" +}