new file: .gitignore

new file: l1_2/hook_mlp_out/cfg.json
new file: l1_2/hook_mlp_out/sae_weights.safetensors
new file: l1_2/hook_mlp_out/sparsity.safetensors
new file: l1_2/hook_resid_post/cfg.json
new file: l1_2/hook_resid_post/sae_weights.safetensors
new file: l1_2/hook_resid_post/sparsity.safetensors
new file: l1_2/hook_z/cfg.json
new file: l1_2/hook_z/sae_weights.safetensors
new file: l1_2/hook_z/sparsity.safetensors

.gitignore

@@ -0,0 +1 @@
+.DS_Store

l1_2/hook_mlp_out/cfg.json

@@ -0,0 +1 @@
+{"model_name": "1l-gelu", "model_class_name": "HookedTransformer", "hook_name": "blocks.0.hook_mlp_out", "hook_eval": "NOT_IN_USE", "hook_layer": 0, "hook_head_index": null, "dataset_path": "ghidav/arithmetics", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 18, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 512, "d_sae": 8192, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": true, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": false, "init_encoder_as_decoder_transpose": false, "n_batches_in_buffer": 128, "training_tokens": 10000000, "finetuning_tokens": 0, "store_batch_size_prompts": 8, "train_batch_size_tokens": 1024, "normalize_activations": "none", "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": false, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 2.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 488, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 1953, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 2000, "dead_feature_window": 1000, "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": "sae-feature-circuits", "wandb_id": null, "run_name": "L0_hook_mlp_out_L1_2_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 0, "checkpoint_path": "checkpoints/137wo4wk", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.18.2", "sae_lens_training_version": "3.18.2", "tokens_per_buffer": 2359296}

l1_2/hook_mlp_out/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:056716c73fedacf71c7d54125f28627d3a8b36a81a57afaeb1575875411edf30
+size 33622400

l1_2/hook_mlp_out/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11dd14aa1f01e235279e3a1758a2e03e654811157f38f53a01e725e4c10e2380
+size 32848

l1_2/hook_resid_post/cfg.json

@@ -0,0 +1 @@
+{"model_name": "1l-gelu", "model_class_name": "HookedTransformer", "hook_name": "blocks.0.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 0, "hook_head_index": null, "dataset_path": "ghidav/arithmetics", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 18, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 512, "d_sae": 8192, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": true, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": false, "init_encoder_as_decoder_transpose": false, "n_batches_in_buffer": 128, "training_tokens": 10000000, "finetuning_tokens": 0, "store_batch_size_prompts": 8, "train_batch_size_tokens": 1024, "normalize_activations": "none", "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": false, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 2.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 488, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 1953, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 2000, "dead_feature_window": 1000, "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": "sae-feature-circuits", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_2_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 0, "checkpoint_path": "checkpoints/ag0mdvt3", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.18.2", "sae_lens_training_version": "3.18.2", "tokens_per_buffer": 2359296}

l1_2/hook_resid_post/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efe3a105dfc0d86300fe672a5424d136c3485e7bc06cffc2cb9897c1b4957755
+size 33622400

l1_2/hook_resid_post/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ab1be493fcbd5d8074a237d66669df4cfa74066dfd6300ca2e28febe1dc9cf0
+size 32848

l1_2/hook_z/cfg.json

@@ -0,0 +1 @@
+{"model_name": "1l-gelu", "model_class_name": "HookedTransformer", "hook_name": "blocks.0.attn.hook_z", "hook_eval": "NOT_IN_USE", "hook_layer": 0, "hook_head_index": null, "dataset_path": "ghidav/arithmetics", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 18, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 512, "d_sae": 8192, "b_dec_init_method": "zeros", "expansion_factor": 16, "activation_fn": "relu", "activation_fn_kwargs": {}, "normalize_sae_decoder": true, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": false, "decoder_heuristic_init": false, "init_encoder_as_decoder_transpose": false, "n_batches_in_buffer": 128, "training_tokens": 10000000, "finetuning_tokens": 0, "store_batch_size_prompts": 8, "train_batch_size_tokens": 1024, "normalize_activations": "none", "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": false, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "adam_beta1": 0, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 2.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 488, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 1953, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 2000, "dead_feature_window": 1000, "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": "sae-feature-circuits", "wandb_id": null, "run_name": "L0_attn.hook_z_L1_2_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 0, "checkpoint_path": "checkpoints/6yndhfi6", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.18.2", "sae_lens_training_version": "3.18.2", "tokens_per_buffer": 2359296}

l1_2/hook_z/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95377a2eb0b4006c5f42d5d41659785fc533818c0a2eb06f81aad20e509ea1df
+size 33622400

l1_2/hook_z/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e423b4c06a1fa95b6ab7e82f2d99f3dae781b80e74e2179d39b68efd9d5122aa
+size 32848