diff --git a/.DS_Store b/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..cab74b406c0e623a5fe29c43df9075136ec2b9e8
Binary files /dev/null and b/.DS_Store differ
diff --git a/L0/.DS_Store b/L0/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..75751c6a4b26b13321c3f9e1c490a05ba0738c14
Binary files /dev/null and b/L0/.DS_Store differ
diff --git a/L0/100M/cfg.json b/L0/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/100M/sae_weights.safetensors b/L0/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c8c651a8ef755e0cdcf1778b45a7fdc90ce8684a
--- /dev/null
+++ b/L0/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fccab3d6a4655d7d9a28070802d5849bd449732c7f094a7a4707c7a21f85a7da
+size 37801344
diff --git a/L0/100M/sparsity.safetensors b/L0/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6403dfe75e467e8969f9384a8c684421097ebddf
--- /dev/null
+++ b/L0/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f9dc061a141bff11c5c258dd875a50b00b010666fdcf6026593179c7673d43a
+size 24656
diff --git a/L0/200M/cfg.json b/L0/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/200M/sae_weights.safetensors b/L0/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8c600bdc7c81cb3571cac1434a585265e6bbada5
--- /dev/null
+++ b/L0/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09c286f10aedce685b6437d15d9c61087f222cc9c6b09838f05d89cb78bd7235
+size 37801344
diff --git a/L0/200M/sparsity.safetensors b/L0/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..81c802ff026d3ddc8588fe1b80bd2f59ad31da7a
--- /dev/null
+++ b/L0/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d21789328ae9fc2a63ec8e8faf24b5dbc19658b13f97baaa385c2ae378cbae1
+size 24656
diff --git a/L0/300M/cfg.json b/L0/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/300M/sae_weights.safetensors b/L0/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..407187668c42fe976580e2a28aa4ff92f3a57ad0
--- /dev/null
+++ b/L0/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee5919165a9355c63f2efefdd6e5019e355a80e3357f3f15b1c704d5f65133d0
+size 37801344
diff --git a/L0/300M/sparsity.safetensors b/L0/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7ff2f31c713fbfb67511799004097439297c113b
--- /dev/null
+++ b/L0/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80be15be93d330fc81936c5c6eae253b3d5f4e2ed36352f3428eba630f43b4e8
+size 24656
diff --git a/L0/400M/cfg.json b/L0/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/400M/sae_weights.safetensors b/L0/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..db11a204465a589096b175eec9735a287e0ec01d
--- /dev/null
+++ b/L0/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d526a6a9271e9f5cbec6eb558bfae653b6a80d83797e574dae29468a315d61cb
+size 37801344
diff --git a/L0/400M/sparsity.safetensors b/L0/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5f8558b72f87eb0a69ac96e463fe113d54267fb6
--- /dev/null
+++ b/L0/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8658d33a95393f5270361d27b3a920901fc1d92b194c6664d34ccc99422940d
+size 24656
diff --git a/L0/500M/cfg.json b/L0/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/500M/sae_weights.safetensors b/L0/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..353bff27c0b95639f00327eaea0439d0d34f1eca
--- /dev/null
+++ b/L0/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3981a91bd59dd240fb183b7d5ddfb976ee0a6c47117f9f8e2c7b2994397ed9c5
+size 37801344
diff --git a/L0/500M/sparsity.safetensors b/L0/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..4a095e3f04f876a4b30561f1076af21132b9a065
--- /dev/null
+++ b/L0/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a5cefff7398f4abfb34b354c4a134ca227d754a690f8cbec435b82e5844d05a
+size 24656
diff --git a/L0/600M/cfg.json b/L0/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/600M/sae_weights.safetensors b/L0/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8105f3d5f24e71323b0e9addb369e7a057b1e83d
--- /dev/null
+++ b/L0/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec985f14abe659420046e2bc17193f9dcb584c117012d7dc189d3f68f7800fb5
+size 37801344
diff --git a/L0/600M/sparsity.safetensors b/L0/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..97dd78fe331f7cc8dd89975ce449b4877822bb28
--- /dev/null
+++ b/L0/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:770c5bef03a7b7360361cc76446a9bc37c36ed23db0b548c2cb149d2da07a441
+size 24656
diff --git a/L0/700M/cfg.json b/L0/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/700M/sae_weights.safetensors b/L0/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0c83b710f9db57b2c940b1d50288715281a2e9e0
--- /dev/null
+++ b/L0/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53057b1bfbdeb962b16248c00808b53b0bd82c31d1e06055ec9828b95a93b6c7
+size 37801344
diff --git a/L0/700M/sparsity.safetensors b/L0/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0a1e557a77bdcc790b3121aa8586f0f9b9428281
--- /dev/null
+++ b/L0/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40446dfac2931e9264cb5a7c0c0e04f190fd290acac4fb42721733fe239377c5
+size 24656
diff --git a/L0/800M/cfg.json b/L0/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/800M/sae_weights.safetensors b/L0/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c7e21dec6b667da152df4d528102a80ae5b19a10
--- /dev/null
+++ b/L0/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d791c8e056991edc50031b4568f00616ebaeb32bd5054ce5f9904d6bb3df3649
+size 37801344
diff --git a/L0/800M/sparsity.safetensors b/L0/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3203caea4362d3d1391a5c6a2fc85643c5ff7a7f
--- /dev/null
+++ b/L0/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bb920eb73d4af9845d2f60a9987609e91b5f65ad52f762b2f39ab3bd4c7b716
+size 24656
diff --git a/L0/900M/cfg.json b/L0/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/900M/sae_weights.safetensors b/L0/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b6bf5dcd7c72a843551c97a58b028a897805e30c
--- /dev/null
+++ b/L0/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:342f2458e8c8a3162605ffe70a14efd59321ab153d2ba02dab30d2b3fbb8ff03
+size 37801344
diff --git a/L0/900M/sparsity.safetensors b/L0/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..bd08e315b6ad8bee5c33d7ee2a082348e2e43741
--- /dev/null
+++ b/L0/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf563f58b90ed8e879518e2b502d41aa0423acdf8ec158364f7737f9c62ae958
+size 24656
diff --git a/L0/cfg.json b/L0/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..c16b9c79f526d79c57c9b51fc788b592a7f8d1ad
--- /dev/null
+++ b/L0/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "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": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L0_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/0gs7wqs5", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L0/sae_weights.safetensors b/L0/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9a1653ab566214545bfc232c51a3c88053742797
--- /dev/null
+++ b/L0/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6acb7f40a9c8dd71edd3742ee221103ede3afb872a06b51ecd4560c1171e1502
+size 37801344
diff --git a/L0/sparsity.safetensors b/L0/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..452d0e385671b95a6f20af44126388399790d099
--- /dev/null
+++ b/L0/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e93cee41bb74dca9fd92ed2659b273a82d61bad27333b9929b8d99493e4b6d10
+size 24656
diff --git a/L1/.DS_Store b/L1/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..b9ee0b08a119c3c5dce21a778b58e3071dc1adfe
Binary files /dev/null and b/L1/.DS_Store differ
diff --git a/L1/100M/cfg.json b/L1/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/100M/sae_weights.safetensors b/L1/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d54dcd14abf6c3e0f856752a2f430adcbad4dc33
--- /dev/null
+++ b/L1/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e78395d5410e0860eed2082fa36ea49dc2faeab0f44791a475ed6ab285c4817f
+size 37801344
diff --git a/L1/100M/sparsity.safetensors b/L1/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..78888c08fcc3552445b173de848a721c3c0dcd46
--- /dev/null
+++ b/L1/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f780a71547cc3b7e730cf7bdb8366a3e142b8a45828fcabe3fb2fd7dfa21f255
+size 24656
diff --git a/L1/200M/cfg.json b/L1/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/200M/sae_weights.safetensors b/L1/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5772640cd04207b21b58753160cb4b36b87a3375
--- /dev/null
+++ b/L1/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76544a19b4eadde706709ea942ea05ce614480de937f0f27597032c6c8906c31
+size 37801344
diff --git a/L1/200M/sparsity.safetensors b/L1/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5acfbfc38df66c971402c956e128d48e2ad5a3b1
--- /dev/null
+++ b/L1/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e49627664d08a1ced5f3517643a662b57e5b09f1dee19a4c6eb8ef921963184
+size 24656
diff --git a/L1/300M/cfg.json b/L1/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/300M/sae_weights.safetensors b/L1/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..53d0255d4d1c5155384ae37250839355cd7a483d
--- /dev/null
+++ b/L1/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7572de6c43b8d37140aa71f743bfd23c184d92e8f171b30bbacbebe995779f4
+size 37801344
diff --git a/L1/300M/sparsity.safetensors b/L1/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5525932bd49077daadb1bba59234af53819e8f9d
--- /dev/null
+++ b/L1/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c79d622b0cc13bff513aabb456229559f75a7e749f50073779f84722ba7e164c
+size 24656
diff --git a/L1/400M/cfg.json b/L1/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/400M/sae_weights.safetensors b/L1/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8e96d6e4acc0ca8fab9ef966aec0db3ce52a685f
--- /dev/null
+++ b/L1/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:139e44c33cf132b5953523ac55c83deaa48a7e7f911efe4a1d6fb21d6c8de8f8
+size 37801344
diff --git a/L1/400M/sparsity.safetensors b/L1/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..49c608622380141891f6567329b7041307e533b7
--- /dev/null
+++ b/L1/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d068f33713c2495a6a8e5ec6e6d769b1d3af7d6c395c6a6a74a2eae3c741730
+size 24656
diff --git a/L1/500M/cfg.json b/L1/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/500M/sae_weights.safetensors b/L1/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9e7d76e128f3a427b94045e9f1703dc50040e567
--- /dev/null
+++ b/L1/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f38f266105cb2215752bc5df84e635b646f418fc9db323315e1d6134cce2396
+size 37801344
diff --git a/L1/500M/sparsity.safetensors b/L1/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..766c7515301f8d503d84b0ee6370c07eafd43f0e
--- /dev/null
+++ b/L1/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f6310aaf90897b3e305cdac83e5c84b1d0cb4f8519ad54a6f3f7abc4119d8b8
+size 24656
diff --git a/L1/600M/cfg.json b/L1/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/600M/sae_weights.safetensors b/L1/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..674aa51128bf7cba9e66aa9d6af00841e3b06c7a
--- /dev/null
+++ b/L1/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a7d92ff32c6c32f8f1914652bab393d84a173cb55549cf80c1b8ccd6c262a25
+size 37801344
diff --git a/L1/600M/sparsity.safetensors b/L1/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..52cebfff7c3f1e74f0db2a2a0a3d01390fbf23b1
--- /dev/null
+++ b/L1/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09e79d101cf501e1ffd6430e55dca29d5941ee24360629ef0b2f24d8304ae712
+size 24656
diff --git a/L1/700M/cfg.json b/L1/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/700M/sae_weights.safetensors b/L1/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7f2e187bad6d7b0c105e5de54af22f65b92e81b8
--- /dev/null
+++ b/L1/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9308afc8bba17455a8fdcc5224eedb4d83b55ee9d5e64f3a23d18e0f48c94199
+size 37801344
diff --git a/L1/700M/sparsity.safetensors b/L1/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c0daa9e70a25a2df4a521fdb867b1360b48774d7
--- /dev/null
+++ b/L1/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:718081e2fe00f8669385ce5aa225277f19f27b17483454ac4915b47cc7433293
+size 24656
diff --git a/L1/800M/cfg.json b/L1/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/800M/sae_weights.safetensors b/L1/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1f8b486c20fcf4d094ecb0fd12f448471204303e
--- /dev/null
+++ b/L1/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e6081550d52e74027e9ce67486653e16f1a5df3fbb13c1b4aad05506065e5a8
+size 37801344
diff --git a/L1/800M/sparsity.safetensors b/L1/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0abccca621019662ab73ed1f24f94fdfa35500be
--- /dev/null
+++ b/L1/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1efe1a006151533fa16d585e29182782488e95e0e52a16f167f4e2bcdf40a9d9
+size 24656
diff --git a/L1/900M/cfg.json b/L1/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/900M/sae_weights.safetensors b/L1/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6d08c73b2516ddeb0a3ec4089d043fdbc3d4f11f
--- /dev/null
+++ b/L1/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ef1b3589c894da970fc99ab8a7337d7c148ed33330f733a951969e557594c23
+size 37801344
diff --git a/L1/900M/sparsity.safetensors b/L1/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..81d7662505a8a6a71e9ba088b7f260f596c480e6
--- /dev/null
+++ b/L1/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba4f17cc41bd352d7fa32620764ed7344484fa7c629a0070c6597e26e7ff0125
+size 24656
diff --git a/L1/cfg.json b/L1/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..11461a82d414eac9208b97504ad9745fd779e94c
--- /dev/null
+++ b/L1/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.1.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 1, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L1_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/eezx69qa", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L1/sae_weights.safetensors b/L1/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..7769f829e4be974b37449a9495a0db17a0d5f405
--- /dev/null
+++ b/L1/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:113867ac12ca66845b57f406a33b7647b4243dcab468b4abad198843f76224c6
+size 37801344
diff --git a/L1/sparsity.safetensors b/L1/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3de73beb6f0c93a9f0ac01a271f9559da0b93383
--- /dev/null
+++ b/L1/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:174b789045f37a36a15dc10cfc160380a4e7422be85e7b46ddea8099fad4b1fc
+size 24656
diff --git a/L10/.DS_Store b/L10/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..5db076953f3ce7dd1c3c74abb8ca169d501375a0
Binary files /dev/null and b/L10/.DS_Store differ
diff --git a/L10/100M/cfg.json b/L10/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/100M/sae_weights.safetensors b/L10/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b9e3edcaac26a9435a537f71f74370310db60be2
--- /dev/null
+++ b/L10/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c6b93a3aec9195432666611269d0e341fee1320ee8363d919b80c673d711a16
+size 37801344
diff --git a/L10/100M/sparsity.safetensors b/L10/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65980423867b602d53c4cc92bf4605a1b648ac26
--- /dev/null
+++ b/L10/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f81087f8255da8f31081215d0a1dc8627e99bcbb7c1d09e900257dbb3b190598
+size 24656
diff --git a/L10/200M/cfg.json b/L10/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/200M/sae_weights.safetensors b/L10/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0076beae4714914c2bd66adb7e503c4949194dea
--- /dev/null
+++ b/L10/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a55b640dfc3aee1b90e7d7b72ea077f723159efe7d82a3734476f1af9b8ed43
+size 37801344
diff --git a/L10/200M/sparsity.safetensors b/L10/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3423197cd993eb7600a83bcd02a37bfb44ca7f6c
--- /dev/null
+++ b/L10/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a45f83fd65b24f0b00d484ecff6578377e4b750296c000f5956a972d9cd98c20
+size 24656
diff --git a/L10/300M/cfg.json b/L10/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/300M/sae_weights.safetensors b/L10/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3d11ed7b878b942d58a4974d14d72db094b360c0
--- /dev/null
+++ b/L10/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4852ccaa0c83f09cbf981492325927ee8dc9a24e7b4ae4321e317e93f5e6f4f5
+size 37801344
diff --git a/L10/300M/sparsity.safetensors b/L10/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e10db2806682a31f104421d6d6634c7a6d85cac8
--- /dev/null
+++ b/L10/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9212da8aa513473726f9ff43dd523a579c8ff53fdafdd7710774a4b67a6eefbe
+size 24656
diff --git a/L10/400M/cfg.json b/L10/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/400M/sae_weights.safetensors b/L10/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..271d3801a27e95d6f17d6035fd19c08cdf2720a2
--- /dev/null
+++ b/L10/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd05b32281149358e2d3379277609dca20236436ec2fbdd705fee708187b62ce
+size 37801344
diff --git a/L10/400M/sparsity.safetensors b/L10/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e22583c9db4153e908c9f4a1d2b6c2b2653ca228
--- /dev/null
+++ b/L10/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a469e9a87d0bd4777e4148c197dabbe7a47b150a6f10559f28bf406d81df7fd
+size 24656
diff --git a/L10/500M/cfg.json b/L10/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/500M/sae_weights.safetensors b/L10/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b2eed7796560ee3e52d47324267ba97c36e9d62f
--- /dev/null
+++ b/L10/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88b0fdd580e192977a19ab5969b32416535a87d45578904a29c26bddb6e4e8fe
+size 37801344
diff --git a/L10/500M/sparsity.safetensors b/L10/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c22f096922dd15d85fbda6419e65e41057df7c50
--- /dev/null
+++ b/L10/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:532a07ba79cb8b84cf00cfcdbdf4d4eabd37f3b1fe522707b66d5b6ea421a5ea
+size 24656
diff --git a/L10/600M/cfg.json b/L10/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/600M/sae_weights.safetensors b/L10/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f2c0beb02512f9e94c324d484bdb2c09d2856e24
--- /dev/null
+++ b/L10/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9809fa9f549522174bd2e617dc22c4f526bd3ef94e98742ad79a04bbfeb19df2
+size 37801344
diff --git a/L10/600M/sparsity.safetensors b/L10/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ea7074d4eb14d10d8a11b08efa267aedeff0fbe3
--- /dev/null
+++ b/L10/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80cb817a2d536e6f3268a606172e223b5cb7aefa2ee5314388e784b491e8c613
+size 24656
diff --git a/L10/700M/cfg.json b/L10/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/700M/sae_weights.safetensors b/L10/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..286eb0c72eea419e42259cf85f1d6e499db95cc7
--- /dev/null
+++ b/L10/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d730d54d97f883bc6c680f3fe170de5197ac5066e055fa1ea07f1031819eb55a
+size 37801344
diff --git a/L10/700M/sparsity.safetensors b/L10/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f0e424e605a8a72b489cbec75c25053ca267f800
--- /dev/null
+++ b/L10/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07864f21d9e7f58727eb09a74309d6ca3cdea9a112843c3699e53b0688db6bc9
+size 24656
diff --git a/L10/800M/cfg.json b/L10/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/800M/sae_weights.safetensors b/L10/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..44675e1bd55a72c555034a24d6b095f5e31d698c
--- /dev/null
+++ b/L10/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20792cb69d4d7f3fd182662c9745c64091a8488a8941167a252e28218008dcf4
+size 37801344
diff --git a/L10/800M/sparsity.safetensors b/L10/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5eeb67300e13503face3bda2bc714bd7959f2f37
--- /dev/null
+++ b/L10/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3f5f05deda220fba4764bdae9d5d07ad960c06d0e2f9835bf4814152a0055a3
+size 24656
diff --git a/L10/900M/cfg.json b/L10/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/900M/sae_weights.safetensors b/L10/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e289589c80bc05963d0b3b6b14ca1d75122f4e6d
--- /dev/null
+++ b/L10/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee3ec22ec0577fcab3081e05643868be105e3f244f000f18209b4d00782b0752
+size 37801344
diff --git a/L10/900M/sparsity.safetensors b/L10/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..db4eaac70fb53c538357b37b3369627007cecddd
--- /dev/null
+++ b/L10/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4eb8b3158e0567dfb79c63759e36715350b70bb911a4405584b5ddedc6c8a06
+size 24656
diff --git a/L10/cfg.json b/L10/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..84b3a4ab909d29dcecaf227ddd906db636ffa256
--- /dev/null
+++ b/L10/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.10.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 10, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L10_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/ty4tmelj", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L10/sae_weights.safetensors b/L10/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..44f2667bf1326fd8462300152925505a8961f8a4
--- /dev/null
+++ b/L10/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca3c7f84228d86a3243162744b2c7d3fbeec208eb0c6fbdba80157c94d8d522d
+size 37801344
diff --git a/L10/sparsity.safetensors b/L10/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..953f7da6e4541363756a31801bc65d40fd486720
--- /dev/null
+++ b/L10/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f819e1d5dab7f01e71151ae043b0bba84d1466cf8ca7246f613236a45e6ed3c
+size 24656
diff --git a/L2/.DS_Store b/L2/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..4ba1bc42b1e8d82e474a7cc553f060c213831ed1
Binary files /dev/null and b/L2/.DS_Store differ
diff --git a/L2/100M/cfg.json b/L2/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/100M/sae_weights.safetensors b/L2/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..faa4a5cdedb7ca5d79bfe69cda09d5d7c3ceee72
--- /dev/null
+++ b/L2/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49a93b10701450e3b0a5e2cb004e144f6b38f776a9cfa84d207886c3f7f458ab
+size 37801344
diff --git a/L2/100M/sparsity.safetensors b/L2/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e8c541db962f80766b7466a1b3620d19dda3068d
--- /dev/null
+++ b/L2/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ac3b10fb23b572543570b33f0ffa180bb6703a6310f5173223e7baabf44f7b5
+size 24656
diff --git a/L2/200M/cfg.json b/L2/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/200M/sae_weights.safetensors b/L2/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a1b4f784ff445b5f203ca49cc48505c1fc895456
--- /dev/null
+++ b/L2/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d546a947a2a8f9d387d110c941a528fcc1dbff421a2e2183ed5b752e67dea5d1
+size 37801344
diff --git a/L2/200M/sparsity.safetensors b/L2/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..75ee56110860c8a249736aa41df6548f49c3384f
--- /dev/null
+++ b/L2/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:35ce315d2f907f8da5f2b4012df38189f39da77e13f69bf6a54c69db9d7d0557
+size 24656
diff --git a/L2/300M/cfg.json b/L2/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/300M/sae_weights.safetensors b/L2/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d1a0d7242df480185ebeeda75b6b51838ca73d38
--- /dev/null
+++ b/L2/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cc28753ebf9d13bcf1ec5e5f62a74f88ed2f22ff6d01be7d43e21e908e957dc
+size 37801344
diff --git a/L2/300M/sparsity.safetensors b/L2/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f3ad11b5753434e5ef9c82d52dd3a077c31c6b7d
--- /dev/null
+++ b/L2/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af225167f332756b4ad21efc6017fec43a1cfb321bbb4a95bad4d9160de031ba
+size 24656
diff --git a/L2/400M/cfg.json b/L2/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/400M/sae_weights.safetensors b/L2/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..19f716f0a28da48956b3289d585778e0da38e639
--- /dev/null
+++ b/L2/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aca9732ecab28289865fdaee8080dd9d44414c76b82678cb03ddd990a97233a5
+size 37801344
diff --git a/L2/400M/sparsity.safetensors b/L2/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..292a3ecd3691ba54c8638d638ea89d409f641c4a
--- /dev/null
+++ b/L2/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b07bad8aa19acc00fa2fa3e4af5229b63ddee7705ec6f83a2b494d88d53cb1fe
+size 24656
diff --git a/L2/500M/cfg.json b/L2/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/500M/sae_weights.safetensors b/L2/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a7e6d2cb004cbcdc7624018a2a6aec50214671be
--- /dev/null
+++ b/L2/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e52dbd4da5f0b76d89d0f7badfc0c20b62a6918ec8c5b43881a8cd75a49455ea
+size 37801344
diff --git a/L2/500M/sparsity.safetensors b/L2/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1e14e010c902123b666e26d0671f1a782d478e1e
--- /dev/null
+++ b/L2/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d7e74be6ece5dc3dc91d56f570b282858b3e97a3ed22af9c0b9ca291867213c
+size 24656
diff --git a/L2/600M/cfg.json b/L2/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/600M/sae_weights.safetensors b/L2/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d30bce5a5b99fb03baa6fcd0aeeaa2dc0a13aeb3
--- /dev/null
+++ b/L2/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cc2ef4ed1bdf6fbc3ef668b035807e5882c4655d11f115b33e2bcbc3fcfad0e
+size 37801344
diff --git a/L2/600M/sparsity.safetensors b/L2/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1968475bda57ef188254d7ee45a0bd6ac87cb75e
--- /dev/null
+++ b/L2/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93c5fcd08964d62ab5c93f967202d03bc00ecdc402f23aa96fa1aeb16387bf7c
+size 24656
diff --git a/L2/700M/cfg.json b/L2/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/700M/sae_weights.safetensors b/L2/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0632307059aa97acfa6286e8fbc8e4168d8c2466
--- /dev/null
+++ b/L2/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:881deceaadbdb1d8f58c64b2d9ae4c5819eaf07e8e5a529ecd7a554cd8fc5aa9
+size 37801344
diff --git a/L2/700M/sparsity.safetensors b/L2/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..357d79548fc94a23ca3dc229c5fcd50e8cffb0b9
--- /dev/null
+++ b/L2/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5c36e4cc5abfb29192769a36d6a778626f817c4bb134943b3386664442d9c5c
+size 24656
diff --git a/L2/800M/cfg.json b/L2/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/800M/sae_weights.safetensors b/L2/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1a8778bf0414c7460aac12ae50becdc866c9efc7
--- /dev/null
+++ b/L2/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64c1abb7ec1f5e5812cd57724f2aa9961ebf7e8c3056d72d490d129fac80ec59
+size 37801344
diff --git a/L2/800M/sparsity.safetensors b/L2/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d9fa048120bafb42aca1701af8007e677288e791
--- /dev/null
+++ b/L2/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e609e7f1932574eb439d24a167cdb969b266373fc61ad52ea362efe00bb9983
+size 24656
diff --git a/L2/900M/cfg.json b/L2/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/900M/sae_weights.safetensors b/L2/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..6faad60bd0399c1e12098b7dad8d210a5697031b
--- /dev/null
+++ b/L2/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f399ede9e1f6442e061d1902d5628012f341f814a81e24796f5b5d9d0b92cd8
+size 37801344
diff --git a/L2/900M/sparsity.safetensors b/L2/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f27a081741a3d98b874d36f8178f917f89c54083
--- /dev/null
+++ b/L2/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84d2fcb3912b500ff4acf526083bf647597222f230e491de435ffd04e66b0259
+size 24656
diff --git a/L2/cfg.json b/L2/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2c2bf2cb50263ab88a1c39355a6ba74719fd8bd4
--- /dev/null
+++ b/L2/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.2.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 2, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L2_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/74k12zkd", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L2/sae_weights.safetensors b/L2/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..676d5ce711f4c3428de53c6aad528c7642d57855
--- /dev/null
+++ b/L2/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1a1f78127fd5ea8761c635166d599b726332bff51e9aa4059e1eb4d34710906
+size 37801344
diff --git a/L2/sparsity.safetensors b/L2/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..845b4ad9e418e2902ee5cc5edfc7e5e23463cda6
--- /dev/null
+++ b/L2/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32b964e6227a44ebab212053051675cd85103082466548731032c33ab91cc482
+size 24656
diff --git a/L4/.DS_Store b/L4/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..d957f0cf022db075e07ec7f2be7ce5eeab3cefcf
Binary files /dev/null and b/L4/.DS_Store differ
diff --git a/L4/100M/cfg.json b/L4/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/100M/sae_weights.safetensors b/L4/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..71ed25e58e8cb1158b39bfd21eab82a2fd0a0b2d
--- /dev/null
+++ b/L4/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6e1217f7ff13a67e1ec5bd4e7239dcaee372ca877ea5fd625c95caec4cd88e9
+size 37801344
diff --git a/L4/100M/sparsity.safetensors b/L4/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f0bbd03ae4ed66397e8199957076a87068a3294b
--- /dev/null
+++ b/L4/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8dee055c6a7bc6d567768d8dafcb08659a4643501467d832e1b1580a5b178f51
+size 24656
diff --git a/L4/200M/cfg.json b/L4/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/200M/sae_weights.safetensors b/L4/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..02836f0320faf5f7f77f258fa09e1e3f80555f18
--- /dev/null
+++ b/L4/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0164940d2f5109f0860b981cc5069932fd6e3e714c27525c91cdbd94945a6d29
+size 37801344
diff --git a/L4/200M/sparsity.safetensors b/L4/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..65e4f04a5710ace8196953c31d40d22aa301d59b
--- /dev/null
+++ b/L4/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f6437224f9402bc71772e8b9082c67763165c0d425a56559cc1048e8388a4f1
+size 24656
diff --git a/L4/300M/cfg.json b/L4/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/300M/sae_weights.safetensors b/L4/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c9064b5f59fffc653fb639c3b044f300964dc7d4
--- /dev/null
+++ b/L4/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e53ca613f4ae54c0f0d6dfaa2a36d44695dc8395a7e6e68d59d91a92449158bb
+size 37801344
diff --git a/L4/300M/sparsity.safetensors b/L4/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..718b4785f954a4c5371aee843a55f274fde52d58
--- /dev/null
+++ b/L4/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dff0b3eadfc64781c95d1bcac55b7eda83a08ad9da6494165a087b0bd3a16fec
+size 24656
diff --git a/L4/400M/cfg.json b/L4/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/400M/sae_weights.safetensors b/L4/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0596b58c19af5641f7fdf16a8cf33bcd18e3c37b
--- /dev/null
+++ b/L4/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c71b14283216175ca06b3117a28eff80c1aff58b9e7406b240d34516d41f7c15
+size 37801344
diff --git a/L4/400M/sparsity.safetensors b/L4/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b68ca1d40ae0dd45ed4a243f08e6a1edd4a38047
--- /dev/null
+++ b/L4/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6fd2c793ff1f4d29b746214d9836ee400ff24d7d0399c5ca7ec257f8c2b109e
+size 24656
diff --git a/L4/500M/cfg.json b/L4/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/500M/sae_weights.safetensors b/L4/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a079e9d5fbb270099f008597ed008ea4df2107c9
--- /dev/null
+++ b/L4/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2094f2f1eede72ca75d267d8f6838fd4d8f50fd12c77f9553cca51c0c1be84a
+size 37801344
diff --git a/L4/500M/sparsity.safetensors b/L4/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c5340fef3e73280f0561cc7f5dba08beb44f5ec3
--- /dev/null
+++ b/L4/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:447f2b2cd207717aba80c4c747979e49709174e01346b690e8f2c07be915518b
+size 24656
diff --git a/L4/600M/cfg.json b/L4/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/600M/sae_weights.safetensors b/L4/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a39335b7b40fbf19bbfd544e0bb4a84cc7a1676d
--- /dev/null
+++ b/L4/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df1435376e36520d1369ca2e4cce309062acb65ce475cbde0f38ef83677878a2
+size 37801344
diff --git a/L4/600M/sparsity.safetensors b/L4/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d79fadd858783e2f4fd2eee8d6aff4bfdf9fca69
--- /dev/null
+++ b/L4/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77f51d6f79aa10edd0fdf0846205588c46f1e50c149a689f171d6c2c7ddb0a6f
+size 24656
diff --git a/L4/700M/cfg.json b/L4/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/700M/sae_weights.safetensors b/L4/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c68ba257dd1d4475a175bca77f337fe2c8db1c82
--- /dev/null
+++ b/L4/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24264f3ec37c6cbb4588c1a04d9a5a02ec9c4b23e436d00a6375d4ba697b2f21
+size 37801344
diff --git a/L4/700M/sparsity.safetensors b/L4/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cc8fe5c2ca5d2bab21ebee54b7123ffaec311316
--- /dev/null
+++ b/L4/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af109c472bfac8d23eb9b90c8e75b1fdcfc22db48e737cbc8c00802bb7da2497
+size 24656
diff --git a/L4/800M/cfg.json b/L4/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/800M/sae_weights.safetensors b/L4/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f43a2a1094c23ee3479181f6a5f9e36d35fb008a
--- /dev/null
+++ b/L4/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b52045b542fe91b43ca182ea09f759e7c2b9e4569cd8a6049f24e8acb26da270
+size 37801344
diff --git a/L4/800M/sparsity.safetensors b/L4/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e9b4417756bcfa6cbbb5fc6d53799919652280be
--- /dev/null
+++ b/L4/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c24873c7a4275375bcebb6c6bf5a4517d588a11394a439fb41375cbc1cd7611
+size 24656
diff --git a/L4/900M/cfg.json b/L4/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/900M/sae_weights.safetensors b/L4/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..629c32342f3d54ff6f503e3cb7d641138c26e4e7
--- /dev/null
+++ b/L4/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:898083d941bbba74097a4c4da012ab0ceefb4d0da97dda70ab263d10de504601
+size 37801344
diff --git a/L4/900M/sparsity.safetensors b/L4/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a0bc070336660e0d489eb6e2a2c810c0defed771
--- /dev/null
+++ b/L4/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec4229f42f56ecebe92c315922f698ec9a3118bf897700d865c3626d9e2832a8
+size 24656
diff --git a/L4/cfg.json b/L4/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..73f963e33c90d05555ffecf2d40e32c9bbecc972
--- /dev/null
+++ b/L4/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.4.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 4, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L4_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/9hlnqk5i", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L4/sae_weights.safetensors b/L4/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..82b12ec75c7f8a11fb8d7051cbbfb02c2e19538d
--- /dev/null
+++ b/L4/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80747516ff91d84590f9f34d7bc8de6f3dd314fca1b763499ac74db90ad76045
+size 37801344
diff --git a/L4/sparsity.safetensors b/L4/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0057a872baef7f4dc3d5dd0045aa03f0ce32eb4b
--- /dev/null
+++ b/L4/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d62798cba6069bc0942c8cfeae560083463b3c9bf8bd675b8a32ee56ee2dc01
+size 24656
diff --git a/L6/.DS_Store b/L6/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..f83290c7ee051c06e7481ec1a5edfe095af98e38
Binary files /dev/null and b/L6/.DS_Store differ
diff --git a/L6/100M/cfg.json b/L6/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/100M/sae_weights.safetensors b/L6/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5d99d8f661d4ce112ad13eede7c310d058d89bed
--- /dev/null
+++ b/L6/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6c13a56cc17dbba10a33621cd76ee30a4db98e2a8791484253706b3e91061b9
+size 37801344
diff --git a/L6/100M/sparsity.safetensors b/L6/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..24d001a6b5f8f0aab3fa0e3104ca3e6938ebc4a7
--- /dev/null
+++ b/L6/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc53d762aefb7fef9cb81a392f140ae09ac007e3e7649d4bfddff56bc97a0db5
+size 24656
diff --git a/L6/200M/cfg.json b/L6/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/200M/sae_weights.safetensors b/L6/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1c14860845c988aaacd0582d97ea11502d1ed41b
--- /dev/null
+++ b/L6/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7027628eec43eca670b124c5d5e30910c98eb181e8449c0bcaf7461ca2e0f762
+size 37801344
diff --git a/L6/200M/sparsity.safetensors b/L6/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..fbd4cadb3730252e1cc751dcf3cca5620ab4d07b
--- /dev/null
+++ b/L6/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:626c6b6400914973c803a8ab2324e461933ea7463ba3fb9b98e425675dd5a184
+size 24656
diff --git a/L6/300M/cfg.json b/L6/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/300M/sae_weights.safetensors b/L6/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0534e1eb094dca310b0bd450ae0c88109c9eba96
--- /dev/null
+++ b/L6/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:213e784dacbe49a0234fb0c52797d9d1faa34a14e7363c7815ab6551e6abaef1
+size 37801344
diff --git a/L6/300M/sparsity.safetensors b/L6/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f31634dc314bf6014ccd05089f4f9d6dfbbefbbf
--- /dev/null
+++ b/L6/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:272681ea23ec1ed2af27dab58969261c22d01bc12127507f81e875faa87ab94b
+size 24656
diff --git a/L6/400M/cfg.json b/L6/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/400M/sae_weights.safetensors b/L6/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b535ac8d5e6267a528862c9d62ec23139a044a95
--- /dev/null
+++ b/L6/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed03a8a2cb3591dfa6aa7c3e223b6948559f964f3c3dcdf75a38560a53d07e40
+size 37801344
diff --git a/L6/400M/sparsity.safetensors b/L6/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..58361901b4342a6e95df859e9ce716303f952401
--- /dev/null
+++ b/L6/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73419d2b06af5682161965b2184aba21d699d450a820ada054e245022fb4c6d0
+size 24656
diff --git a/L6/500M/cfg.json b/L6/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/500M/sae_weights.safetensors b/L6/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..57f8cb9b11a483726fd63ef65fa4f4f6a7d74449
--- /dev/null
+++ b/L6/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6be26d50a5eaa87e47783d1fc9d15609226a1d6f086a4a0b287aed7a44f74d69
+size 37801344
diff --git a/L6/500M/sparsity.safetensors b/L6/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..3d7fd6130e1c768d4aad18d6f6401e00a65ef8a0
--- /dev/null
+++ b/L6/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2b6c0fcd7125e8de8f357448527dd3bb6c8e04f1d222b046a1b902ff850c659
+size 24656
diff --git a/L6/600M/cfg.json b/L6/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/600M/sae_weights.safetensors b/L6/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5a587b701210a28ef4bd56ac2e3c87b19d057eec
--- /dev/null
+++ b/L6/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:058ffd9befe1dd5a91d6ce95a8eb476107d14adeb5cfd802245c7e3dc53cc858
+size 37801344
diff --git a/L6/600M/sparsity.safetensors b/L6/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..5c417411fbb114907e0f2cb7a6bc3c9cb168c89d
--- /dev/null
+++ b/L6/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:268e643a78f09260eaf7ef79b1febed669d436606ec6158a6c02d0acdd1f771d
+size 24656
diff --git a/L6/700M/cfg.json b/L6/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/700M/sae_weights.safetensors b/L6/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f536caea1e29705da38a4eccddfc31f9a478fb11
--- /dev/null
+++ b/L6/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b93299eee20b1b36e0989b8414a25660afa6c2093fd5a905dcbacc55a5fa53
+size 37801344
diff --git a/L6/700M/sparsity.safetensors b/L6/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..337d649899958a8e41c1ba4ba395d7ef1d89d85c
--- /dev/null
+++ b/L6/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce5a3be69c986b3a1f387c02ebfd2b203ca618f3c3533af573e8e889a2dc0944
+size 24656
diff --git a/L6/800M/cfg.json b/L6/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/800M/sae_weights.safetensors b/L6/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..12f85b29a6df28ff810f58de36d7fa93f371c6be
--- /dev/null
+++ b/L6/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c28249d1dd16b9049e0fa360ef131ab0ab003a58947d8c095d2dabb269431a47
+size 37801344
diff --git a/L6/800M/sparsity.safetensors b/L6/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f8ef2a643b0159edbb0c39a948b5e96252a7439f
--- /dev/null
+++ b/L6/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1467a83245dbb079891046950fb0b9a07b65c606405b4f541a593ffba979bb09
+size 24656
diff --git a/L6/900M/cfg.json b/L6/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/900M/sae_weights.safetensors b/L6/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a23d7ce14b6c3ac8b086d017b58e70e47b83dca6
--- /dev/null
+++ b/L6/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9706bd31b8adb9c774b0d7cea6a523f0bd07a25f24756a7137de0586118508a3
+size 37801344
diff --git a/L6/900M/sparsity.safetensors b/L6/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e9c556ba9053b269e88ba7a51c79556ebf82009b
--- /dev/null
+++ b/L6/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3aed8f3ab736bf726791704d3f7b0d5487169c53e60a5386983134aadbec4033
+size 24656
diff --git a/L6/cfg.json b/L6/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..2e332c204908e25bf1fac87222bf5bc4561dc462
--- /dev/null
+++ b/L6/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.6.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 6, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L6_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/4bppuj63", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L6/sae_weights.safetensors b/L6/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d268a26e0b57a8947d8c9b79a42ad3d3eaf97e35
--- /dev/null
+++ b/L6/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a793ded70ec91c54449df4a1519fcc0f641372f6dbd94faeb077759ed2dc0a6
+size 37801344
diff --git a/L6/sparsity.safetensors b/L6/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..a6f1c1851d8f92a879b0b1209180bc6a441c509a
--- /dev/null
+++ b/L6/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2b8476f4468db77d4c02adc6fe2a90dca09d07630ecc9e9df78d3794fe59b7f
+size 24656
diff --git a/L8/.DS_Store b/L8/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..c00baaf1c36d5a7c6243dc2e50bd4fcb81bdfb81
Binary files /dev/null and b/L8/.DS_Store differ
diff --git a/L8/100M/cfg.json b/L8/100M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/100M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/100M/sae_weights.safetensors b/L8/100M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..88cc4845955b2bfe3742ff783dff27b500fe0667
--- /dev/null
+++ b/L8/100M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3846df920c3342b46b8265f68dd272a985fe6f8e7695f88241cfdbcf95455783
+size 37801344
diff --git a/L8/100M/sparsity.safetensors b/L8/100M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..843a56a86719205593da984af0159d9f1e18e052
--- /dev/null
+++ b/L8/100M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80ba8e13e8b4e7a63d96cd5db80b1a6709d02b93fb11ba4ff4e8daf3977fbef0
+size 24656
diff --git a/L8/200M/cfg.json b/L8/200M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/200M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/200M/sae_weights.safetensors b/L8/200M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..8463d91913bdd0b0b5f6fd2a3f0976e47de04158
--- /dev/null
+++ b/L8/200M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74c89a49694cf953d2d9c43ac25e84dac90e9ce3a075685080f655dbb310ba8d
+size 37801344
diff --git a/L8/200M/sparsity.safetensors b/L8/200M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d81458972af01e4a9b16371f2cf7e929223e2a0e
--- /dev/null
+++ b/L8/200M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3083b373bb20f8726cc4edc274cf508954cf07b3c80d42df0a078949ad2add7
+size 24656
diff --git a/L8/300M/cfg.json b/L8/300M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/300M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/300M/sae_weights.safetensors b/L8/300M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..e4613ead6171d9ce92a025391f5e2116f936aca1
--- /dev/null
+++ b/L8/300M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02d2776071cddefbf6650363636a2b717001d635a84e0a7534324d1dd2c07667
+size 37801344
diff --git a/L8/300M/sparsity.safetensors b/L8/300M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c25b8a68b0c756890c13f9b8d226c48368a7ce36
--- /dev/null
+++ b/L8/300M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f9c241364c598360f2f03b534542b86747c666d1a71a605134300a52004708b
+size 24656
diff --git a/L8/400M/cfg.json b/L8/400M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/400M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/400M/sae_weights.safetensors b/L8/400M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..9876ab5d4547abd8993fcf5baca67697d118a245
--- /dev/null
+++ b/L8/400M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:135ad2fa235b314e0f28adb41c44a21e1daed74c4a28405dc3014a84bfff664a
+size 37801344
diff --git a/L8/400M/sparsity.safetensors b/L8/400M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..14620be46152320f69ecbbb7b2123d2b1ba3137f
--- /dev/null
+++ b/L8/400M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b63507fe580e883a1cd94117149c2ee11391cb6a621e284b5ec4443ce287aab
+size 24656
diff --git a/L8/500M/cfg.json b/L8/500M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/500M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/500M/sae_weights.safetensors b/L8/500M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1043281c9cbd06e76f08de999c37dd3d28a9331a
--- /dev/null
+++ b/L8/500M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b31156cd4943d1f70c10186399aedcce60ef88d113e86397ed69e03c15f105a6
+size 37801344
diff --git a/L8/500M/sparsity.safetensors b/L8/500M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..853c9e54175aee59710057a83fc8da7bb2119b87
--- /dev/null
+++ b/L8/500M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d24e51f8b1cbc3286b96c402c0d301f3d3d5b370f5ebef3e0ad9e358c2345751
+size 24656
diff --git a/L8/600M/cfg.json b/L8/600M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/600M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/600M/sae_weights.safetensors b/L8/600M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c021c26ff15a1089ab712e66ecaa8e297ac9b855
--- /dev/null
+++ b/L8/600M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe1d2ea2f72633d3f41e1b4a2a3011626e886a61ed712659c1e817e9558506d8
+size 37801344
diff --git a/L8/600M/sparsity.safetensors b/L8/600M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c0d6d551eb2e4ef6da94e075d7ce195ae91cfa72
--- /dev/null
+++ b/L8/600M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:492c9aeb8625e577366951767042a68f2f06bb257e56fc4b97e21e8ad27366fc
+size 24656
diff --git a/L8/700M/cfg.json b/L8/700M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/700M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/700M/sae_weights.safetensors b/L8/700M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..63c7fdb7abc1ef72d2e35662ee94768364751138
--- /dev/null
+++ b/L8/700M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:add8cfda4e17559612060bedff267b625886b337ce460537b255f4dea3a74523
+size 37801344
diff --git a/L8/700M/sparsity.safetensors b/L8/700M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..f927667cefdb557504775dd60a9bd0c37a9fe4e6
--- /dev/null
+++ b/L8/700M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca47ac78b67fc65f7e84db0934ae3c584f1f279cf53e5fdcccfc48109063c1ab
+size 24656
diff --git a/L8/800M/cfg.json b/L8/800M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/800M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/800M/sae_weights.safetensors b/L8/800M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..1cffd64edb2df7f4a857aaec6d8e4b3238af7b72
--- /dev/null
+++ b/L8/800M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:737bc1fccd01eb88260624b631ce030c721ca1ca24d48099bb97a46e06303ca0
+size 37801344
diff --git a/L8/800M/sparsity.safetensors b/L8/800M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..ef3f58b1b8786c98ab41021863cc65b69e3dd049
--- /dev/null
+++ b/L8/800M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:472dceb7e9a3f7524ef8f071915ebaf76d97db01bd006690c985a7f1eb9852fe
+size 24656
diff --git a/L8/900M/cfg.json b/L8/900M/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/900M/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/900M/sae_weights.safetensors b/L8/900M/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..0a5f90cb427fe1730dea3eccb5b3a947bd166357
--- /dev/null
+++ b/L8/900M/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b609b1bf1c08d957042e7a0c565bd9434529c54b88aca7b90db82a148ea76b7
+size 37801344
diff --git a/L8/900M/sparsity.safetensors b/L8/900M/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..058ad00e08df0cf0ec94ebb1e3f1f0ba33af4def
--- /dev/null
+++ b/L8/900M/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c520ee40b689340ff6553ee6fe5689c0da8efca7ccf00d8827ae8d7a77d8116
+size 24656
diff --git a/L8/cfg.json b/L8/cfg.json
new file mode 100644
index 0000000000000000000000000000000000000000..38da2908e298eed19ea5eb9e5e503086766320b7
--- /dev/null
+++ b/L8/cfg.json
@@ -0,0 +1 @@
+{"model_name": "pythia-160m-deduped", "model_class_name": "HookedTransformer", "hook_name": "blocks.8.hook_resid_post", "hook_eval": "NOT_IN_USE", "hook_layer": 8, "hook_head_index": null, "dataset_path": "NeelNanda/pile-small-tokenized-2b", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": true, "context_size": 1024, "use_cached_activations": false, "cached_activations_path": null, "architecture": "jumprelu", "d_in": 768, "d_sae": 6144, "b_dec_init_method": "zeros", "expansion_factor": 8, "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": 1000000000, "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": 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": 1.0, "lp_norm": 1, "scale_sparsity_penalty_by_decoder_norm": false, "l1_warm_up_steps": 12207, "lr": 3e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 3e-06, "lr_decay_steps": 48828, "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-transfer-learning", "wandb_id": null, "run_name": "L8_hook_resid_post_L1_1_0", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 100, "resume": false, "n_checkpoints": 10, "checkpoint_path": "checkpoints/bet5fk1u", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {}, "sae_lens_version": "3.13.1", "sae_lens_training_version": "3.13.1", "tokens_per_buffer": 536870912}
\ No newline at end of file
diff --git a/L8/sae_weights.safetensors b/L8/sae_weights.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..c59924aeba519bf5dcdda388a90cb8a51c576e81
--- /dev/null
+++ b/L8/sae_weights.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d961b088585640ff8e344424dd5d60b8790336c4b575347b8c5e7d3b662937aa
+size 37801344
diff --git a/L8/sparsity.safetensors b/L8/sparsity.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..509c6d9bee76b11682205b3880b6f242abb86ce0
--- /dev/null
+++ b/L8/sparsity.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51cbe8a4dea526ae66d3c6fe5d3b155a998981a6b964230b6b83be327b0e157b
+size 24656