Upload SAE blocks.4.hook_mlp_out

blocks.4.hook_mlp_out/cfg.json

@@ -1 +1 @@
-{"model_name": "gpt2-small", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_mlp_out", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 256, "use_cached_activations": false, "cached_activations_path": null, "architecture": "topk", "d_in": 768, "d_sae": 24576, "b_dec_init_method": "zeros", "expansion_factor": 32, "activation_fn": "topk", "activation_fn_kwargs": {"k": 32}, "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": 64, "training_tokens": 40960000, "finetuning_tokens": 0, "store_batch_size_prompts": 16, "normalize_activations": "expected_average_only_in", "seqpos_slice": [null], "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": true, "jumprelu_init_threshold": 0.001, "jumprelu_bandwidth": 0.001, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "train_batch_size_tokens": 4096, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 0.001, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 0, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 2000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "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": "sae_lens_tutorial", "wandb_id": null, "run_name": "24576-L1-0.001-LR-5e-05-Tokens-4.096e+07", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 1, "checkpoint_path": "checkpoints/ot3n85a0", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"center_writing_weights": false}, "sae_lens_version": "5.9.1", "sae_lens_training_version": "5.9.1", "exclude_special_tokens": false, "tokens_per_buffer": 67108864}
+{"model_name": "gpt2-small", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_mlp_out", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "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": 256, "use_cached_activations": false, "cached_activations_path": null, "architecture": "topk", "d_in": 768, "d_sae": 12288, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "topk", "activation_fn_kwargs": {"k": 100}, "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, "decoder_heuristic_init_norm": 0.1, "init_encoder_as_decoder_transpose": true, "n_batches_in_buffer": 64, "training_tokens": 40960000, "finetuning_tokens": 0, "store_batch_size_prompts": 16, "normalize_activations": "expected_average_only_in", "seqpos_slice": [null], "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": true, "jumprelu_init_threshold": 0.001, "jumprelu_bandwidth": 0.001, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "train_batch_size_tokens": 4096, "adam_beta1": 0.9, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 0.001, "lp_norm": 1.0, "scale_sparsity_penalty_by_decoder_norm": true, "l1_warm_up_steps": 0, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 2000, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 1000, "dead_feature_window": 1000, "dead_feature_threshold": 0.0001, "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": "gpt2small-mlp-out-saes", "wandb_id": null, "run_name": "12288-L1-0.001-LR-5e-05-Tokens-4.096e+07", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20, "resume": false, "n_checkpoints": 0, "checkpoint_path": "checkpoints/ho1chtgt", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"center_writing_weights": false}, "sae_lens_version": "5.10.1", "sae_lens_training_version": "5.10.1", "exclude_special_tokens": false, "tokens_per_buffer": 67108864}

blocks.4.hook_mlp_out/sae_weights.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ff51f67c353d98d4d22f96d8b10479eaedea8da40017014f00a084773d7cb2b
-size 151096640
+oid sha256:ff9d63737462cf24815ce71af5ae16083daba7d4a01e0d87414a07e9348208f8
+size 75550008

blocks.4.hook_mlp_out/sparsity.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5982e050845ba901c7059978a16b62feb873e0013e9ef23c0e7000b9e764c677
-size 98384
+oid sha256:7915a6646b957fdf5259e62ee7cb330810a414353db1e320f2a88e9c75cc4689
+size 49232