c1park commited on
Commit
0894ad5
1 Parent(s): 0f011d2

Upload folder using huggingface_hub

Browse files
sae_gpt2-small_blocks.8.hook_resid_pre_12288:v0/cfg.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"model_name": "gpt2-small", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_pre", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "apollo-research/Skylion007-openwebtext-tokenizer-gpt2", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "standard", "d_in": 768, "d_sae": 12288, "b_dec_init_method": "geometric_median", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "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": 128, "training_tokens": 122880000, "finetuning_tokens": 0, "store_batch_size_prompts": 32, "train_batch_size_tokens": 4096, "normalize_activations": "none", "device": "cuda:2", "act_store_device": "cuda:2", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "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": 1500, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 0, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 25000, "dead_feature_window": 5000, "dead_feature_threshold": 1e-08, "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": "how_we_train_SAEs_replication_1", "wandb_id": null, "run_name": "12288-L1-2-LR-5e-05-Tokens-1.229e+08", "wandb_entity": null, "wandb_log_frequency": 100, "eval_every_n_wandb_logs": 10, "resume": false, "n_checkpoints": 0, "checkpoint_path": "checkpoints/w90bfj8v", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.12.0", "sae_lens_training_version": "3.12.0", "tokens_per_buffer": 536870912}
sae_gpt2-small_blocks.8.hook_resid_pre_12288:v0/sae_weights.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a34c044e10d8e67e15941ae985cfc0b4b174ec8a88a4bb5dc6c400ba43ffabc
3
+ size 75550008
sae_gpt2-small_blocks.8.hook_resid_pre_12288:v0/sparsity.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5e1c5eda914fceb51fc3a3c2686d02c589f995430fac6c5c9e5a928e20e7a359
3
+ size 49232