Upload Gemma-2b IT SAE

gemma_2b_it_blocks.12.hook_resid_post_16384/cfg.json

@@ -0,0 +1 @@
+{"model_name": "gemma-2b-it", "model_class_name": "HookedTransformer", "hook_name": "blocks.12.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 12, "hook_head_index": null, "dataset_path": "chanind/openwebtext-gemma", "dataset_trust_remote_code": true, "streaming": false, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "d_in": 2048, "d_sae": 16384, "b_dec_init_method": "zeros", "expansion_factor": 8, "activation_fn": "relu", "normalize_sae_decoder": false, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": true, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 16, "training_tokens": 1228800000, "finetuning_tokens": 0, "store_batch_size_prompts": 8, "train_batch_size_tokens": 4096, "normalize_activations": "none", "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": true, "llm_compilation_mode": "max-autotune", "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 2, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 15000, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 60000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 5000, "dead_feature_window": 5000, "dead_feature_threshold": 1e-06, "n_eval_batches": 10, "eval_batch_size_prompts": null, "log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "gemma_2b_exploration", "wandb_id": null, "run_name": "16384-L1-2-LR-5e-05-Tokens-1.229e+09", "wandb_entity": null, "wandb_log_frequency": 50, "eval_every_n_wandb_logs": 10, "resume": false, "n_checkpoints": 5, "checkpoint_path": "checkpoints/adk2ig0r", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.5.0", "sae_lens_training_version": "3.5.0", "tokens_per_buffer": 67108864}

gemma_2b_it_blocks.12.hook_resid_post_16384/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:702d6e94fd9740ceffabf86045f8147a3c128093d3333b15c6122df3d19fea66
+size 268509504

gemma_2b_it_blocks.12.hook_resid_post_16384/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f499057b339d8006b2b6852868cb74823cef83eb8e3d79544b66f53b455e5148
+size 65616