| { | |
| "_name_or_path": "/data5/models/Qwen2-57B-A14B-Instruct", | |
| "architectures": [ | |
| "Qwen2MoeForCausalLM" | |
| ], | |
| "attention_dropout": 0.0, | |
| "bos_token_id": 151643, | |
| "decoder_sparse_step": 1, | |
| "eos_token_id": 151643, | |
| "hidden_act": "silu", | |
| "hidden_size": 3584, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 18944, | |
| "max_position_embeddings": 32768, | |
| "max_window_layers": 28, | |
| "mlp_only_layers": [], | |
| "model_type": "qwen2_moe", | |
| "moe_intermediate_size": 2560, | |
| "norm_topk_prob": false, | |
| "num_attention_heads": 28, | |
| "num_experts": 64, | |
| "num_experts_per_tok": 8, | |
| "num_hidden_layers": 28, | |
| "num_key_value_heads": 4, | |
| "output_router_logits": false, | |
| "quantization_config": { | |
| "amp": true, | |
| "autoround_version": "0.3.0.dev", | |
| "backend": "auto_round:exllamav2", | |
| "bits": 4, | |
| "data_type": "int", | |
| "dataset": "NeelNanda/pile-10k", | |
| "enable_minmax_tuning": true, | |
| "enable_quanted_input": true, | |
| "extra_config": { | |
| "model.layers.3.mlp.shared_expert.down_proj": { | |
| "clip": true | |
| }, | |
| "model.layers.0.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.0.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.1.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.1.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.10.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.10.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.11.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.11.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.12.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.12.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.13.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.13.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.14.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.14.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.15.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.15.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.16.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.16.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.17.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.17.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.18.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.18.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.19.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.19.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.2.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.2.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.20.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.20.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.21.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.21.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.22.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.22.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.23.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.23.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.24.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.24.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.25.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.25.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.26.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.26.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.27.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.27.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.3.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.3.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.4.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.4.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.5.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.5.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.6.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.6.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.7.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.7.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.8.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.8.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.9.mlp.gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| }, | |
| "model.layers.9.mlp.shared_expert_gate": { | |
| "bits": 32, | |
| "data_type": "bfloat", | |
| "group_size": null, | |
| "sym": null | |
| } | |
| }, | |
| "gradient_accumulate_steps": 1, | |
| "group_size": 128, | |
| "iters": 1000, | |
| "low_gpu_mem_usage": false, | |
| "lr": 0.001, | |
| "minmax_lr": 0.001, | |
| "nsamples": 512, | |
| "quant_method": "intel/auto-round", | |
| "scale_dtype": "torch.float16", | |
| "seqlen": 2048, | |
| "sym": false, | |
| "train_bs": 8 | |
| }, | |
| "rms_norm_eps": 1e-06, | |
| "rope_theta": 1000000.0, | |
| "router_aux_loss_coef": 0.001, | |
| "shared_expert_intermediate_size": 20480, | |
| "sliding_window": 65536, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "bfloat16", | |
| "transformers_version": "4.41.1", | |
| "use_cache": true, | |
| "use_sliding_window": false, | |
| "vocab_size": 151936 | |
| } | |