diff --git a/k32-sae-32k-seed2/config.json b/k32-sae-32k-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..f6a3f6d8a91318454732084d28245595acbed5d3 --- /dev/null +++ b/k32-sae-32k-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32767, "k": 32, "multi_topk": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-32k-seed-2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-32k-seed2/layers.6/cfg.json b/k32-sae-32k-seed2/layers.6/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..fc0a090df0744d87a2d0fc7c27a6d36d7b6ef2a6 --- /dev/null +++ b/k32-sae-32k-seed2/layers.6/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32767, "k": 32, "multi_topk": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-32k-seed2/layers.6/sae.safetensors b/k32-sae-32k-seed2/layers.6/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..550ed39e38b00264c03a528fc977be270fd63c21 --- /dev/null +++ b/k32-sae-32k-seed2/layers.6/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16b40cd627dd5d4b46033cd36f30707bb0abb2f16b2dcee5c9d1ffda7c256220 +size 201454924 diff --git a/k32-sae-32k-seed2/lr_scheduler.pt b/k32-sae-32k-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..78112e643f40c8c3a9bc10ee31295cf5cb4a4009 --- /dev/null +++ b/k32-sae-32k-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f0a1e452458dcff0a7d7c508004ac9dd6172452eab713e5cf8e037254b55ff1 +size 1012 diff --git a/k32-sae-32k-seed2/optimizer.pt b/k32-sae-32k-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..7a68f91ab061db9fe7e654213911d1f13f18ada6 --- /dev/null +++ b/k32-sae-32k-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a8d536df4f49670ebb57f98551e2c773c9261ddd17b84fb3377db87cd17ae41 +size 102313038 diff --git a/k32-sae-32k-seed2/state.pt b/k32-sae-32k-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..6e42dee9e30c0d282c7710993bee532d6a622f4e --- /dev/null +++ b/k32-sae-32k-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01f44192946218fe46687d074f00646485b9caaafcf4278c59eed86882fba20a +size 263314 diff --git a/k32-sae-32k-seed3/config.json b/k32-sae-32k-seed3/config.json new file mode 100644 index 0000000000000000000000000000000000000000..74fca757aff7b4beffd65cfd0a94c604bdc68731 --- /dev/null +++ b/k32-sae-32k-seed3/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32k-seed3", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 52, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-32k-seed3/layers.6/cfg.json b/k32-sae-32k-seed3/layers.6/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-32k-seed3/layers.6/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-32k-seed3/layers.6/sae.safetensors b/k32-sae-32k-seed3/layers.6/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f48a617b1d19112c2cd1166d7edf576d8e06110c --- /dev/null +++ b/k32-sae-32k-seed3/layers.6/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1eaca564411d77eb066645d7d70c90a9ab8cf2af3daf1a580d8db1508578f42c +size 201461072 diff --git a/k32-sae-32k-seed3/lr_scheduler.pt b/k32-sae-32k-seed3/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..09d44e04064f4b6efd4e262a530222ffda2bae63 --- /dev/null +++ b/k32-sae-32k-seed3/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a645210450394a692a612289cb5fe097161dfa420f3c634f6516bc67841ac2b4 +size 1012 diff --git a/k32-sae-32k-seed3/optimizer.pt b/k32-sae-32k-seed3/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..5d1c39b2d0ad0a2c13ee834412a70014cfa81390 --- /dev/null +++ b/k32-sae-32k-seed3/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c76608cb52470444eff72adfb2856858648c97fca54c139888d5cd90f23ca92 +size 102316366 diff --git a/k32-sae-32k-seed3/state.pt b/k32-sae-32k-seed3/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..2b52bc37887fc9d887537d2ca43ce12d9efc22d5 --- /dev/null +++ b/k32-sae-32k-seed3/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:030d953fc60b9fade27a7f202235308c1b5910d15c53c99ff0153648d3f7efd2 +size 263314 diff --git a/k32-sae-4k-seed2/config.json b/k32-sae-4k-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..12b85e4f98eee409ae41babf74a68fbdbdbf1b0c --- /dev/null +++ b/k32-sae-4k-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-4k-seed-2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-4k-seed2/layers.6/cfg.json b/k32-sae-4k-seed2/layers.6/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..c1a15e209466e9f749442cdb2fa281cac6612cdf --- /dev/null +++ b/k32-sae-4k-seed2/layers.6/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-4k-seed2/layers.6/sae.safetensors b/k32-sae-4k-seed2/layers.6/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f1ea53e7cf1633b2614416c4113219cf6bca545a --- /dev/null +++ b/k32-sae-4k-seed2/layers.6/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10f3b782c57111d9b70bddf462b9f3b2ddce9f2acf8e5f1903b6ec8ae3003aa7 +size 25111832 diff --git a/k32-sae-4k-seed2/lr_scheduler.pt b/k32-sae-4k-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..3f886e7b57bf9f42f53236df1084c98e70b138a4 --- /dev/null +++ b/k32-sae-4k-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee7d6621abae43ef404e05f222fac18c5b0579dd18d6dfa06ce767eb4da83c84 +size 1012 diff --git a/k32-sae-4k-seed2/optimizer.pt b/k32-sae-4k-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..cf508004d0dd6efb98e5abdeffb48459e52b5c49 --- /dev/null +++ b/k32-sae-4k-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1da6314ae5fe35c78c7c2a9f8c5d90a3e2fcb826b867ad49b15a3df848c79ada +size 12787664 diff --git a/k32-sae-4k-seed2/state.pt b/k32-sae-4k-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..976087464806631b208a01ee0fe820b5c450b1e0 --- /dev/null +++ b/k32-sae-4k-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8efed45dd478ea04f30ecad126e8e39188dc200a94c8fe8e4f9ac5cc3f119e0 +size 33874 diff --git a/k32-sae-4k/config.json b/k32-sae-4k/config.json new file mode 100644 index 0000000000000000000000000000000000000000..3dbe05a251edebd14a0bfe70c5a99598d2294396 --- /dev/null +++ b/k32-sae-4k/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-4k", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-4k/layers.6/cfg.json b/k32-sae-4k/layers.6/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..c1a15e209466e9f749442cdb2fa281cac6612cdf --- /dev/null +++ b/k32-sae-4k/layers.6/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-4k/layers.6/sae.safetensors b/k32-sae-4k/layers.6/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4c91e656aa0272c8138b03c340d288d65382ccf1 --- /dev/null +++ b/k32-sae-4k/layers.6/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff14a1d338e11b6ca41a490ac9342b21ac28e597e49348cfad88a335efd3bdee +size 25111832 diff --git a/k32-sae-4k/lr_scheduler.pt b/k32-sae-4k/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..3f886e7b57bf9f42f53236df1084c98e70b138a4 --- /dev/null +++ b/k32-sae-4k/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee7d6621abae43ef404e05f222fac18c5b0579dd18d6dfa06ce767eb4da83c84 +size 1012 diff --git a/k32-sae-4k/optimizer.pt b/k32-sae-4k/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..0f99ef5f11269e5a43c4afec24fa2b134e1f55cf --- /dev/null +++ b/k32-sae-4k/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e859e10167a20db5e4a7b30d086194a0f07e6fb384b41992ba04118f57d3aec9 +size 12787664 diff --git a/k32-sae-4k/state.pt b/k32-sae-4k/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..6be966787a15d82d5bc46b03a7e3831726afeb46 --- /dev/null +++ b/k32-sae-4k/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0322e5e343f9de89481204249014b0cf21e1574df473ebb1101fd826e10ec30a +size 33874 diff --git a/k32-sae-mlp-131k-seed2/config.json b/k32-sae-mlp-131k-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..1475821bdb522934fbc429c15a9b43f3423b4cac --- /dev/null +++ b/k32-sae-mlp-131k-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 131072, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-131k-seed2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 22, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-131k-seed2/layers.6.mlp/cfg.json b/k32-sae-mlp-131k-seed2/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..558e52f7536ee3df1fb0bc06f5c07a999475a77b --- /dev/null +++ b/k32-sae-mlp-131k-seed2/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 131072, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-131k-seed2/layers.6.mlp/sae.safetensors b/k32-sae-mlp-131k-seed2/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1efa343536c17dab02a936c424a4cc2d10a0adb4 --- /dev/null +++ b/k32-sae-mlp-131k-seed2/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4221255f0e0ab6d83d5a0ca2a82057cc136e548b7ece89a3f9fc10ff16917b8 +size 805834064 diff --git a/k32-sae-mlp-131k-seed2/lr_scheduler.pt b/k32-sae-mlp-131k-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..f16140c160b877ce6bf86f1953ad5b884a846845 --- /dev/null +++ b/k32-sae-mlp-131k-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:362ac3a1db985156bc8176ba7ca7e961052de2b14537e00920c9d0ced3332f83 +size 1012 diff --git a/k32-sae-mlp-131k-seed2/optimizer.pt b/k32-sae-mlp-131k-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..2c2172b542eea6a7c83f9d52efcd0b23d4658028 --- /dev/null +++ b/k32-sae-mlp-131k-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:399683a7d9f55b264a13ce603ba3bc208056d2b42445fbf30379b2d4604f958a +size 409224590 diff --git a/k32-sae-mlp-131k-seed2/state.pt b/k32-sae-mlp-131k-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..3d4d74f0ecfc561be613549269e3bdf455a60dbb --- /dev/null +++ b/k32-sae-mlp-131k-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:702086fe6e3933ee528e639b59c34fa6050e7aa481538fdf531ef0625d5d23b6 +size 1049746 diff --git a/k32-sae-mlp-131k/config.json b/k32-sae-mlp-131k/config.json new file mode 100644 index 0000000000000000000000000000000000000000..328de67734464a980b1257bbd52e2aeae6aa1d88 --- /dev/null +++ b/k32-sae-mlp-131k/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 131072, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-131k", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-131k/layers.6.mlp/cfg.json b/k32-sae-mlp-131k/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..558e52f7536ee3df1fb0bc06f5c07a999475a77b --- /dev/null +++ b/k32-sae-mlp-131k/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 131072, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-131k/layers.6.mlp/sae.safetensors b/k32-sae-mlp-131k/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5c20ab8d74ba67bfc335354a4799fe2170f008a1 --- /dev/null +++ b/k32-sae-mlp-131k/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f227709d7296ee81b47df96e823ae05cad6896cb200296a98f15c7e6e5112570 +size 805834064 diff --git a/k32-sae-mlp-131k/lr_scheduler.pt b/k32-sae-mlp-131k/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..f16140c160b877ce6bf86f1953ad5b884a846845 --- /dev/null +++ b/k32-sae-mlp-131k/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:362ac3a1db985156bc8176ba7ca7e961052de2b14537e00920c9d0ced3332f83 +size 1012 diff --git a/k32-sae-mlp-131k/optimizer.pt b/k32-sae-mlp-131k/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..8e362976b8a95baf32593b98231b956fc553e37e --- /dev/null +++ b/k32-sae-mlp-131k/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af46175f5379285fbd13a4202c53bdf9058453445d1e014b2807c2c93c0f8af1 +size 409224590 diff --git a/k32-sae-mlp-131k/state.pt b/k32-sae-mlp-131k/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..1ecb73ee58bc21b1125156b4824f562eee64673a --- /dev/null +++ b/k32-sae-mlp-131k/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:315e73d5973c4fe4362f3e7ab6db5e111279736148ff3a66b30665db9d4d33ce +size 1049746 diff --git a/k32-sae-mlp-32-seed2/config.json b/k32-sae-mlp-32-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..2e750617773fe762d4172ccaeb0a42f876467d33 --- /dev/null +++ b/k32-sae-mlp-32-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32-seed2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 22, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.0.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.0.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..eecec2350d73e25d5e8ceaefad2a481ace58685a --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e38ba482665bf6cf883ef2ae7eb4c03fe58eeed7c70512cde2145e1a09aa9338 +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.1.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.1.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4a569d761477a8436b7100d6bfebc9e29052c34d --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a3a60c4f4fe47161150cd5a9eec22c0af508a60628d657e532fecfa9a3317d16 +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.10.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.10.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2e507a3d9d12bea96f629ba762bf33fa01108000 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:630bd111b8ba1156a167ddbc1362f07ef69c9254e0b68ba4f859fb1692a2261f +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.11.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.11.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..088e028beeebd62827e70b42c9e45bc138cb9da8 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fc5f01c1902724ba4df2f5f4e29e4798ae102af6c1c70972ef7adfc402e37ab +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.2.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.2.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..394c7f4443a9660c069210831837bd0fbf283919 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd808d179120860cef241ec92c0f2598f7926dcc88263749f42ce3b27494b9ea +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.3.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.3.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..22d5e62ae7bebaf214b9f1c7d922edb97da8c396 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10f6e149b0ba5f2ec371e722eb6379dad50a3d89d985b7d6621edaa7e6f42599 +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.4.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.4.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2199c9b5e00e9013889a80d9d853dddb0761693a --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7889a1742f0d0696b88d07445f9e66c9b974d586d7896f02359bdfc38306b90a +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.5.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.5.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3d093a6bc3bd5dab50cb9a916079ec67e3a06395 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:68af5c0fdd99f12a9305e440b91d0e1315bfa2dcdcabc04ec76345a95c827876 +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.6.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.6.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b484c35cd1b0a605432af88b6beeddbc5d1a5089 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0546caa33f17cd905ac0b43d924fa249563a0b4ade682f67d6827e2c76eefdf +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.7.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.7.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4664f53b166109b84c9b5052ec909492e324cd91 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41c993338912c7682f2b2e03183637e3268b75c8e15add9c5171bdb03f2fb62d +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.8.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.8.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b3423a625024b324eddf5d20424382e4ecc6e63e --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28b9749079ddc58ed4ad2194c249d611d8521c81c06526b159220e85afbe0ca5 +size 200112 diff --git a/k32-sae-mlp-32-seed2/layers.9.mlp/cfg.json b/k32-sae-mlp-32-seed2/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32-seed2/layers.9.mlp/sae.safetensors b/k32-sae-mlp-32-seed2/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..694ca773852756177edd90a9584d1ae1f1f94844 --- /dev/null +++ b/k32-sae-mlp-32-seed2/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b45f848fac2c3e951dc405b47cca71d442f2b03f927098b81d03ee045be71c0 +size 200112 diff --git a/k32-sae-mlp-32-seed2/lr_scheduler.pt b/k32-sae-mlp-32-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..85fc1a065aa1a38f23d3935886323d43e15597f8 --- /dev/null +++ b/k32-sae-mlp-32-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c6a3f7fa4b7a769daff54c693b7d5117f0f3335aeab7a1e14bedb14129e0509f +size 1268 diff --git a/k32-sae-mlp-32-seed2/optimizer.pt b/k32-sae-mlp-32-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..5ffb5a4a8ed59e5a8fb900f22fc984ab2ec38dcc --- /dev/null +++ b/k32-sae-mlp-32-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59f858958c832ea11eb4d461516a64243edff5990c9ffea30ebd57785b658740 +size 1320826 diff --git a/k32-sae-mlp-32-seed2/state.pt b/k32-sae-mlp-32-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..59b93afddaa38eb5b8a20aa59fb7ab396711291a --- /dev/null +++ b/k32-sae-mlp-32-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afee5d18f3fb7226baa5c86a12712e5434aebec77eb8ac0ec7ac1acb32b505fb +size 6482 diff --git a/k32-sae-mlp-32/config.json b/k32-sae-mlp-32/config.json new file mode 100644 index 0000000000000000000000000000000000000000..7851726a91be942eae2a8ce0c809e6b4e4ef1e9c --- /dev/null +++ b/k32-sae-mlp-32/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.0.mlp/cfg.json b/k32-sae-mlp-32/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.0.mlp/sae.safetensors b/k32-sae-mlp-32/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c87f16926924ca62898ccfcb03d793bf378fd3f6 --- /dev/null +++ b/k32-sae-mlp-32/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acfbdd2521e2b44735d092201f40228e55d875798acb44d16afff9b3bc5fcf32 +size 200112 diff --git a/k32-sae-mlp-32/layers.1.mlp/cfg.json b/k32-sae-mlp-32/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.1.mlp/sae.safetensors b/k32-sae-mlp-32/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a2246a61709d93c165555ec04b4c31fd098cd372 --- /dev/null +++ b/k32-sae-mlp-32/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cdf92d8df019fe42b13067946628c83fd619ea60a0383664ab395ce25f23823b +size 200112 diff --git a/k32-sae-mlp-32/layers.10.mlp/cfg.json b/k32-sae-mlp-32/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.10.mlp/sae.safetensors b/k32-sae-mlp-32/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..38341f44248452a9c81fe14c33330b0d96bb79e5 --- /dev/null +++ b/k32-sae-mlp-32/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97f65f8aa46d85043e7b9214177c81ddfa246d39449de0fd24890e91c63c774f +size 200112 diff --git a/k32-sae-mlp-32/layers.11.mlp/cfg.json b/k32-sae-mlp-32/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.11.mlp/sae.safetensors b/k32-sae-mlp-32/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b4a1f4215f3d9559fbbda7479cf409b68146c0e5 --- /dev/null +++ b/k32-sae-mlp-32/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd5268169ce84afad8430b3c5427253a19dcbdc99ec6fc95961709cb4ab8bd29 +size 200112 diff --git a/k32-sae-mlp-32/layers.2.mlp/cfg.json b/k32-sae-mlp-32/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.2.mlp/sae.safetensors b/k32-sae-mlp-32/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6cbe0d7f3b0b75d3245ac6b63a230fdc91906f17 --- /dev/null +++ b/k32-sae-mlp-32/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1f40dc209805a2f171f9ab86d9f85748c199bf46f4c47993b49a3293e13bdbc +size 200112 diff --git a/k32-sae-mlp-32/layers.3.mlp/cfg.json b/k32-sae-mlp-32/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.3.mlp/sae.safetensors b/k32-sae-mlp-32/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..52699deef7b8faff9d5c97a437b75604015ceff4 --- /dev/null +++ b/k32-sae-mlp-32/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bd792430dd3b3dd0594840e838d8980a5269eee2a202483899038e3a85e87be +size 200112 diff --git a/k32-sae-mlp-32/layers.4.mlp/cfg.json b/k32-sae-mlp-32/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.4.mlp/sae.safetensors b/k32-sae-mlp-32/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1449f3c51f05cf2b9dadd1e81172faf3271b19b7 --- /dev/null +++ b/k32-sae-mlp-32/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6eff85c05e9ca59ea9e341a8f4acaff20f944b046edcbec0f94211473c5d2d6f +size 200112 diff --git a/k32-sae-mlp-32/layers.5.mlp/cfg.json b/k32-sae-mlp-32/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.5.mlp/sae.safetensors b/k32-sae-mlp-32/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..42a4a6ff099be3ee706c9762807f6e17fc71865f --- /dev/null +++ b/k32-sae-mlp-32/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72a55cb5b484a8e4ed93fe1b902d1a7131cbb98425dfe8f403a73c455a8fe027 +size 200112 diff --git a/k32-sae-mlp-32/layers.6.mlp/cfg.json b/k32-sae-mlp-32/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.6.mlp/sae.safetensors b/k32-sae-mlp-32/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b42a5cac2c9a8c5a349fa135d4da209afb53b82c --- /dev/null +++ b/k32-sae-mlp-32/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43bdbdf7622bcb9b74a2cf064d59425c1640ef7b86829d1fb586d49ccc9e5363 +size 200112 diff --git a/k32-sae-mlp-32/layers.7.mlp/cfg.json b/k32-sae-mlp-32/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.7.mlp/sae.safetensors b/k32-sae-mlp-32/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..638b32eb283c36e42a9bb4851278859a626be3b5 --- /dev/null +++ b/k32-sae-mlp-32/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d00e55c85f02a4df190a3fe96db72749840c1c1884b81ec790b8e7678487b6e7 +size 200112 diff --git a/k32-sae-mlp-32/layers.8.mlp/cfg.json b/k32-sae-mlp-32/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.8.mlp/sae.safetensors b/k32-sae-mlp-32/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ab30c6dcf2f7fb54cb32e3d42c047e8e849be730 --- /dev/null +++ b/k32-sae-mlp-32/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1758240402ca46d511227b00e9134e841b83c36c29fe2d33e17a948036d8df6a +size 200112 diff --git a/k32-sae-mlp-32/layers.9.mlp/cfg.json b/k32-sae-mlp-32/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..93d64302c1b2cc94927710d1d032187dfbf787dc --- /dev/null +++ b/k32-sae-mlp-32/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32/layers.9.mlp/sae.safetensors b/k32-sae-mlp-32/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bdd09c2d0c86edf72e944b1336fda3d13e722855 --- /dev/null +++ b/k32-sae-mlp-32/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4aad3d788e8d2e65b5d51a624f4905c93141ac3f9525a5fa328660064e66183 +size 200112 diff --git a/k32-sae-mlp-32/lr_scheduler.pt b/k32-sae-mlp-32/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..85fc1a065aa1a38f23d3935886323d43e15597f8 --- /dev/null +++ b/k32-sae-mlp-32/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c6a3f7fa4b7a769daff54c693b7d5117f0f3335aeab7a1e14bedb14129e0509f +size 1268 diff --git a/k32-sae-mlp-32/optimizer.pt b/k32-sae-mlp-32/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..0f483194006821e44d48ac137f07d98eceda9bbb --- /dev/null +++ b/k32-sae-mlp-32/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a2eaf67d2c7eff2216d9d051485b8503078c9cbd142c6efd2c777ba63c78396 +size 1320826 diff --git a/k32-sae-mlp-32/state.pt b/k32-sae-mlp-32/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..59b93afddaa38eb5b8a20aa59fb7ab396711291a --- /dev/null +++ b/k32-sae-mlp-32/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afee5d18f3fb7226baa5c86a12712e5434aebec77eb8ac0ec7ac1acb32b505fb +size 6482 diff --git a/k32-sae-mlp-32k-seed2/config.json b/k32-sae-mlp-32k-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..1923a6b36490369dbb8c89c2ba0e177ca320ce51 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32k-seed2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.0.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.0.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1bf27b6e34f782075f2bd485f6e4ba46fe306770 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39ee4e7227168ef0725e1f57d04e66d60f804c9655170aed9fdad0f5fc7234a6 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.1.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.1.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d45b300cec6b49f01236f7dbc25fa6518cedb2e8 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f626727223ec8d1ff70fc5ce6a6ed3af3f71640e0aea95eb64719de34cdf0298 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.10.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.10.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c431fe6b24feadd1a8ae7e83962e8fb4a5e2a70e --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:708983880cb6e4a782f2e0fe0dc6caa93d54fc7ab3e58c134dca0f1e925f035a +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.11.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.11.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f76e2f7187bcf9206023c42b795095811f1b25f5 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a235808bdbd78f7d381c9dd1ca9b72eb93153835c2ad6516f972508a88327158 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.2.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.2.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a2c0ca378792f09db5678131bf69293840b514a4 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2ce00b5f137dfb77461513970369b43410510e97e223a6037b63a4946f87a87 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.3.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.3.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..86fa8c52ce8e08d97dd1b55468ab455f25438508 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a759328ebf4fcd7290d25228045d6d2200ecb5fb6c3c1ffa61833b65ba3161c +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.4.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.4.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f260055634d258960fb272287325e5d678fd351f --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:213a108f78b103b9aec6d2cd8877acfef825f4dbd63a8e05e87b3d3c52e9ffbf +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.5.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.5.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f673f284600fa96997540a1045d063434cfeac95 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3da99221f4ac0ae2bf9c2112fafccb705c82b96913fd9004084601dfa1137823 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.6.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.6.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..033a92fef81a3a127c0bc1bce8ca2ed85abfb9a2 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3df961c33d5204c49da822ee5d3283e6f1f577fcba2076666b52d532df732a52 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.7.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.7.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c4b09de4b2a1da489c8b0208a2ce4c3e517c079b --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6884be5b13f184750ba3f1a0588c0db377a116d5612f5c763a71ff050e827ab +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.8.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.8.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4543687be533bb3a389eed8993c565bbc6ed212f --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c174a061d4c4cafabccd189cc1c243558ece8d3d8ec34858d926ab593bdcbfb2 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/layers.9.mlp/cfg.json b/k32-sae-mlp-32k-seed2/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed2/layers.9.mlp/sae.safetensors b/k32-sae-mlp-32k-seed2/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..685b1eacb60441b360070a5d2ac284fd93dd26ef --- /dev/null +++ b/k32-sae-mlp-32k-seed2/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88e244f65369aca3bac001687f0f0752305005f3786bd6d6973ff3b2fdc82c58 +size 201461072 diff --git a/k32-sae-mlp-32k-seed2/lr_scheduler.pt b/k32-sae-mlp-32k-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..7470aa92e348fac4910c3e581c958ac53c69d7f4 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ee1262d16ccd79f52b27e666309ee963bdc5eff5684b34eb9d272420f13070b +size 1268 diff --git a/k32-sae-mlp-32k-seed2/optimizer.pt b/k32-sae-mlp-32k-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..71595b9601b2f7707b8823c856cf935e77b47df7 --- /dev/null +++ b/k32-sae-mlp-32k-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e05bafe2454f1edc8daf75759ea94fd2bc634a81c0590507e9b0ff079115dd6 +size 1227759482 diff --git a/k32-sae-mlp-32k-seed2/state.pt b/k32-sae-mlp-32k-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..2316c7554e6b43c5a549ce786e2d7ad53e2ab12e --- /dev/null +++ b/k32-sae-mlp-32k-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a8e581d864519f11b65dd5698587660be4cb8a4d914e553bd174a7e0705951d +size 3149138 diff --git a/k32-sae-mlp-32k-seed3/config.json b/k32-sae-mlp-32k-seed3/config.json new file mode 100644 index 0000000000000000000000000000000000000000..ec9ed68d9010334fb5d1609fd296d9b7b75faddf --- /dev/null +++ b/k32-sae-mlp-32k-seed3/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32k-seed3", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 52, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed3/layers.6.mlp/cfg.json b/k32-sae-mlp-32k-seed3/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..ba51bc4beefd90615925438ff54756a7c3161145 --- /dev/null +++ b/k32-sae-mlp-32k-seed3/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-seed3/layers.6.mlp/sae.safetensors b/k32-sae-mlp-32k-seed3/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cf7314876bcf8068ee70eac2fd98c145217c913a --- /dev/null +++ b/k32-sae-mlp-32k-seed3/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ec7c8add04286ffe5865c742bc841dac765e946bd55ab465d6bda2beeea3a63 +size 201461072 diff --git a/k32-sae-mlp-32k-seed3/lr_scheduler.pt b/k32-sae-mlp-32k-seed3/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..10665724254e7c22a7a1404d8f7fb5c40849f60e --- /dev/null +++ b/k32-sae-mlp-32k-seed3/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11b5192fa27cf7089cdfbc43a42ead0e86f1257c5babcb88951ac1987b6c5dbf +size 1012 diff --git a/k32-sae-mlp-32k-seed3/optimizer.pt b/k32-sae-mlp-32k-seed3/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..f080b79c8d6c39f844c14e8c0194b50c26d39267 --- /dev/null +++ b/k32-sae-mlp-32k-seed3/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d4f923bf2445b699c905a9104f88c726484606c5fd5cb1fe43e3a79d0e8e226 +size 102316366 diff --git a/k32-sae-mlp-32k-seed3/state.pt b/k32-sae-mlp-32k-seed3/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..9cb821d79827d0ded7f45d22e4db52a78ba5b3fd --- /dev/null +++ b/k32-sae-mlp-32k-seed3/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f59861d2107118a34be3f2f6846765e1437ea3812ea7ed34e849aa1890bfa3d +size 263314 diff --git a/k32-sae-mlp-32k-tied-seed2/config.json b/k32-sae-mlp-32k-tied-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..43971682d029f0ad2a3fcf415e7f1d234e76bff9 --- /dev/null +++ b/k32-sae-mlp-32k-tied-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32767, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32k-tied-seed2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 22, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-32k-tied-seed2/layers.6.mlp/cfg.json b/k32-sae-mlp-32k-tied-seed2/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..67792f201e0a982504bd64fa96864ed9a0cb2904 --- /dev/null +++ b/k32-sae-mlp-32k-tied-seed2/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32767, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-tied-seed2/layers.6.mlp/sae.safetensors b/k32-sae-mlp-32k-tied-seed2/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4a83bc15d0a1bf20421457e987eb6adf2a1da56f --- /dev/null +++ b/k32-sae-mlp-32k-tied-seed2/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf968b1a8ea5b014ed80b5f2d75f3fa7d29d98fc86a748a5e69b9cc66d980b1a +size 201454924 diff --git a/k32-sae-mlp-32k-tied-seed2/lr_scheduler.pt b/k32-sae-mlp-32k-tied-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..78112e643f40c8c3a9bc10ee31295cf5cb4a4009 --- /dev/null +++ b/k32-sae-mlp-32k-tied-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f0a1e452458dcff0a7d7c508004ac9dd6172452eab713e5cf8e037254b55ff1 +size 1012 diff --git a/k32-sae-mlp-32k-tied-seed2/optimizer.pt b/k32-sae-mlp-32k-tied-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..7cc09918fddf48da4299cf0817295d7175274bc5 --- /dev/null +++ b/k32-sae-mlp-32k-tied-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3e7713a8f2568f7fd9a77585d69b70aacc5d9d64b1704afa09a373d50a49862 +size 51195538 diff --git a/k32-sae-mlp-32k-tied-seed2/state.pt b/k32-sae-mlp-32k-tied-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..a5fbfed5e77c7b01adc4542cc0af045c0211f7fd --- /dev/null +++ b/k32-sae-mlp-32k-tied-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ccb0932259a440d5ebc59fb8e7189e78c679f2442c32b2353540639091e64954 +size 263314 diff --git a/k32-sae-mlp-32k-tied/config.json b/k32-sae-mlp-32k-tied/config.json new file mode 100644 index 0000000000000000000000000000000000000000..81a03c3483e4926eb6c83f8170bd05800e73a5ea --- /dev/null +++ b/k32-sae-mlp-32k-tied/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32767, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.6.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-32k-tied", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-32k-tied/layers.6.mlp/cfg.json b/k32-sae-mlp-32k-tied/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..67792f201e0a982504bd64fa96864ed9a0cb2904 --- /dev/null +++ b/k32-sae-mlp-32k-tied/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 32767, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-32k-tied/layers.6.mlp/sae.safetensors b/k32-sae-mlp-32k-tied/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8465a23ba833521f6a5831d30cc2e539986cdce4 --- /dev/null +++ b/k32-sae-mlp-32k-tied/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51b9a833ada4b175fd97570b688330231c15e6c60c6c213d1d120c42fd0eeffd +size 201454924 diff --git a/k32-sae-mlp-32k-tied/lr_scheduler.pt b/k32-sae-mlp-32k-tied/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..78112e643f40c8c3a9bc10ee31295cf5cb4a4009 --- /dev/null +++ b/k32-sae-mlp-32k-tied/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f0a1e452458dcff0a7d7c508004ac9dd6172452eab713e5cf8e037254b55ff1 +size 1012 diff --git a/k32-sae-mlp-32k-tied/optimizer.pt b/k32-sae-mlp-32k-tied/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..2b50e3b374b26ca3db83bb596519f7868e3ce694 --- /dev/null +++ b/k32-sae-mlp-32k-tied/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50f12ea935a4fc6e235ba86a3e6aa1767672f6264cc664d388e3387cb2159291 +size 51195538 diff --git a/k32-sae-mlp-32k-tied/state.pt b/k32-sae-mlp-32k-tied/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..6c650d703609a4e486910a6329af52d74e0f6e3a --- /dev/null +++ b/k32-sae-mlp-32k-tied/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1d90a95b1c35b73b4547c9f4aff8642db8748348103a4c49f8c7816a409c60f +size 263314 diff --git a/k32-sae-mlp-4k-seed2/config.json b/k32-sae-mlp-4k-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..66fc30a0dcf75e09c32dc2be5bcac92dc47ee1b3 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-4k-seed2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 22, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.0.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.0.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..694b0156b3b9c1ddbec74c16756f9aa7d532613c --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae8b62422e607d3357dc8fcb09608fc33dc9992500b82049e0d898794d4692ed +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.1.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.1.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e4c512dafaba1a919b7150746f5a234c1590d875 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04a434fd577bdbbaac97d5106b4957812a3d504634cd68834d3e094fa3a2a8ab +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.10.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.10.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a2380e77c0744b975f5ec996c14fac62c4ed9f58 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e30a5c1cdd079f092785886e99c9d19f0e4fd7cc9f3137b44fe002ea3c87909 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.11.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.11.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..bf2305aa8351e51a912764cea63fe32c099660b3 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f9f98f4a1c725b557bbe8524e6c98a48d13816f13af123cff6206deb4e92815 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.2.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.2.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7a0c259e06b65c613b04a3ff335752186449af75 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:532ac796e7ce1e9ffd310fc96c0c0674864e5b943773a7c0b8ff8544562d2724 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.3.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.3.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..79936a5636a49b70d8a0876e5854d6a23994d8a8 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ed3276f48362f5b83a7e67fb497048947823fc66a8cc11174704292ed759323 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.4.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.4.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c6625ef11648e784c7b11852af42e97bb0581ae6 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48f4f83555a565459d59a7d3e94cf975477a42bf20372488f4016c9089ff6108 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.5.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.5.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..64bb99bbaf6f585977d3343923ca4903ca44e02e --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5321b4726b86767b7d98e77a210dc9e1a7959c4d03e796962777a59dd5bb6fbd +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.6.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.6.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fdc083bead9728f90057a9f845e289f25f736bd1 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d81ec06ba8076133654b59254fa5224eb8f2560669d2c456dc61f90bfed9f550 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.7.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.7.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b98af7349b4b7229007bb3fb08503c47203eafb9 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be226a5e158eca4a1b5aa28b2503b5aba6be9eeb0c92c8fdd27123813380fd84 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.8.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.8.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7760c6fcf0933f0414a8ca4ba6d53bdd86cc6aff --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f56085c7bf142915458e1a3831641cc159da02c7d8b1d6f4fcd9e54c4d4e12c5 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/layers.9.mlp/cfg.json b/k32-sae-mlp-4k-seed2/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k-seed2/layers.9.mlp/sae.safetensors b/k32-sae-mlp-4k-seed2/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4377781fc857fe721da66a2dd4ba5032320694cd --- /dev/null +++ b/k32-sae-mlp-4k-seed2/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9625637fa1dba8618b1382eb5518d6d54e6db1c957e0eec23b54fc6b560ed814 +size 25111832 diff --git a/k32-sae-mlp-4k-seed2/lr_scheduler.pt b/k32-sae-mlp-4k-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..4592dc50e90e241aed7f9e5e702da7c0753e82be --- /dev/null +++ b/k32-sae-mlp-4k-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e8379fa6dfc2f18e3b78c92820cd1367aeed450447c39a7403aedd37c18676b +size 1268 diff --git a/k32-sae-mlp-4k-seed2/optimizer.pt b/k32-sae-mlp-4k-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..270fe1ea0674b399f85f7c4aab7d56b05cf416dc --- /dev/null +++ b/k32-sae-mlp-4k-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:832181692a4507b0ed9cdf4d13b701f9af2fb4c61569d9f5caa29f0fc27b0b3e +size 153414266 diff --git a/k32-sae-mlp-4k-seed2/state.pt b/k32-sae-mlp-4k-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..cc6cc04a46f980384a7ae53dd145a3da43082f08 --- /dev/null +++ b/k32-sae-mlp-4k-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:159c8db000fe85366e7244b52937e214d833973370a15dc6c32f227673fefb83 +size 395858 diff --git a/k32-sae-mlp-4k/config.json b/k32-sae-mlp-4k/config.json new file mode 100644 index 0000000000000000000000000000000000000000..67247406194a8fef41d00afb56bd1a7af8853154 --- /dev/null +++ b/k32-sae-mlp-4k/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-4k", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.0.mlp/cfg.json b/k32-sae-mlp-4k/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.0.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..8aa49f95cbc68c0dc688bb2d77ee8697dcbaa1dc --- /dev/null +++ b/k32-sae-mlp-4k/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:510f53d90c1699688ef5b9cf6d34c5579f3642c614e3f8cf0bf95871155a528c +size 25111832 diff --git a/k32-sae-mlp-4k/layers.1.mlp/cfg.json b/k32-sae-mlp-4k/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.1.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dcc24d10277c9518738da64215334881d0404488 --- /dev/null +++ b/k32-sae-mlp-4k/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d227618b2ba621e262770a444be326b68fd9da995d743667ce507e06ade615c0 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.10.mlp/cfg.json b/k32-sae-mlp-4k/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.10.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5c3b52fa04ecf97bae3c7422f27cf744e614cd24 --- /dev/null +++ b/k32-sae-mlp-4k/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b047c8cef71e33e6c7973bec764ed25ef9d037acea57fb22e0e619e919df0d1 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.11.mlp/cfg.json b/k32-sae-mlp-4k/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.11.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..13613efc153230de820121b7e39a7886ed5a0dc4 --- /dev/null +++ b/k32-sae-mlp-4k/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41e47643f5d0d26f5cd203b432c564a9d242cf6edd6ad1bcf6bd057dbd94105c +size 25111832 diff --git a/k32-sae-mlp-4k/layers.2.mlp/cfg.json b/k32-sae-mlp-4k/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.2.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..00ae4efa786cc71b94fd103988acd45b0c8b3b26 --- /dev/null +++ b/k32-sae-mlp-4k/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa21db5f6cdb667fab82964044b1f30f1c4b1e83e71db5538e75afe764778f07 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.3.mlp/cfg.json b/k32-sae-mlp-4k/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.3.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1b3ec7c3868126f9d4eb39a2a7ea674a8b25fa61 --- /dev/null +++ b/k32-sae-mlp-4k/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3afa70529bb9b6b9009b9ece4cd710adb86c271bec82615913175489d3b2312b +size 25111832 diff --git a/k32-sae-mlp-4k/layers.4.mlp/cfg.json b/k32-sae-mlp-4k/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.4.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..02f45ee9d80ab1b88917787365f19f0f5e4bc10e --- /dev/null +++ b/k32-sae-mlp-4k/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c760dece1f30b71b9c22d6db819f64a64af64756830e7dc2a6da3e143afecc4 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.5.mlp/cfg.json b/k32-sae-mlp-4k/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.5.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2e9f001f6e5de5ee2a1520425b857572820ce94a --- /dev/null +++ b/k32-sae-mlp-4k/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81dcc3c2de87a83552098569c50f70f4f2532f3fee79f175b5981b4ec5117214 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.6.mlp/cfg.json b/k32-sae-mlp-4k/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.6.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..e9780c1eb141d1c406cdd1b716cc55608c0a2014 --- /dev/null +++ b/k32-sae-mlp-4k/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fde6c5639c2e4b15403959095c80cf82bad268b26c6ecc5b8f2b4ad9a42986f +size 25111832 diff --git a/k32-sae-mlp-4k/layers.7.mlp/cfg.json b/k32-sae-mlp-4k/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.7.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0e84434072eb2f1cbb75c4763b6dedf333f764c2 --- /dev/null +++ b/k32-sae-mlp-4k/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddfdb5c2ca7788489521bee9b7b33013b96f69917d1f073b06062740646816d3 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.8.mlp/cfg.json b/k32-sae-mlp-4k/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.8.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1b06194d191148490e51939a361c5f81b4d6ee8f --- /dev/null +++ b/k32-sae-mlp-4k/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73d6efd5956dc61326d6836c260d5525a4ee538ae7e3ace6bfa3e3ddd2f70e41 +size 25111832 diff --git a/k32-sae-mlp-4k/layers.9.mlp/cfg.json b/k32-sae-mlp-4k/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..20c184fe049979882e5ffcf426e5a9e443d0163d --- /dev/null +++ b/k32-sae-mlp-4k/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 4084, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-4k/layers.9.mlp/sae.safetensors b/k32-sae-mlp-4k/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..832c5d910b8c5a16649a34bb3be23dd745c5a687 --- /dev/null +++ b/k32-sae-mlp-4k/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83dfbf972591aee2a7a74be1254f7015071a377aec486a8ff38d8b2c85a5516a +size 25111832 diff --git a/k32-sae-mlp-4k/lr_scheduler.pt b/k32-sae-mlp-4k/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..4592dc50e90e241aed7f9e5e702da7c0753e82be --- /dev/null +++ b/k32-sae-mlp-4k/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e8379fa6dfc2f18e3b78c92820cd1367aeed450447c39a7403aedd37c18676b +size 1268 diff --git a/k32-sae-mlp-4k/optimizer.pt b/k32-sae-mlp-4k/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..7e93168c37d40a7287edd0acf4af0ef712c2531d --- /dev/null +++ b/k32-sae-mlp-4k/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:983811462ad8bf294169db66bf1865fe1b7b2d3be0596039cbfad7228c2386a6 +size 153414266 diff --git a/k32-sae-mlp-4k/state.pt b/k32-sae-mlp-4k/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..9893f48374e7bf7f10fdbbda3fdf5cc91fa21295 --- /dev/null +++ b/k32-sae-mlp-4k/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01b5982db2cc7358ecc6920603a30284a1f754cdeed7e1c82f629835e431ab76 +size 395858 diff --git a/k32-sae-mlp-768-seed2/config.json b/k32-sae-mlp-768-seed2/config.json new file mode 100644 index 0000000000000000000000000000000000000000..155c273371456f171dd255db0f930bbdca893593 --- /dev/null +++ b/k32-sae-mlp-768-seed2/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-768-seed2", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 22, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.0.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.0.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..50312a303b10c9ea1ec0003798ce763d34d8add8 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2515e6f7687322c88540be549483eeffc1eb8900b39edf7aa10ab2dcb7fe050 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.1.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.1.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..25279ffb7d6a0a9f98bc7ba36caeaa5b0c754f30 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:322d3fa1bd4b21ce7826765f24e8d17b19f4aef248e235f964e05d5100b76c53 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.10.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.10.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c6bd9aa7492b2a297fd7bff20cdceef1e77275ce --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4c72847404760a4fbcb0b6b848548915c2301d73f7161fffad1a5db0ed1fdd3 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.11.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.11.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5889c1215782d7a7d44a85b065d81899156ca0ad --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:54b71591100f71300da26bfc8d4d884bbe71ed5a0baaed90aba5daee57f928e0 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.2.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.2.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1cad33a3defc96f615e6af26f5a96e77c5eac54b --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e3d70c2c873d9b4b5b4df90804809e9c401081e0afad833fb70b3f81287f4f4 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.3.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.3.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c8bc18ab8a8016f34cf97d75c88efad73121acd6 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:836598fd3e4a0ef8b1e3a7f3e4ffd5e9c0073f64ca002771d010839091e57475 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.4.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.4.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..7641da0d0f9c63455cd266c1b129c5ff5de0e0ff --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53d17294fe530c719b2062e50502bb33e35c015f3f2e74b04a834e21ca3a90f6 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.5.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.5.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1868d86c80bf03885160315dc238791afb00b02a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67bd6754a4c43520574a0bcfd69f1307b348bad0918e278653c3b2f28a1fcf7d +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.6.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.6.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cb278e7e05d6e4889f6d6020b6c995ad67f83ab3 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c96e1328c8fd4e527a77fb8e64590ee75cc50239e67bbe1c967475a38adb17f +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.7.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.7.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5c8d93b39a253be206162f1e45505b98e49866a8 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2465e3ffd090bafd11aa19d0896ce09e88897950004639d5a2f6a59c17abbcd4 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.8.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.8.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..de927de58303069e42ee209a0db9ba37c3d42f29 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d927d3ad52f5e37551e6f9924adc64a92b7f67d421968298c72165d14fd9642 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/layers.9.mlp/cfg.json b/k32-sae-mlp-768-seed2/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768-seed2/layers.9.mlp/sae.safetensors b/k32-sae-mlp-768-seed2/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d1e0a059326850ff2635abb334aa365737a5ef73 --- /dev/null +++ b/k32-sae-mlp-768-seed2/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b3c860d9251e192b4362914e5cd9fd246e1e711232ce8406a38b266ab1615f3 +size 4725048 diff --git a/k32-sae-mlp-768-seed2/lr_scheduler.pt b/k32-sae-mlp-768-seed2/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..18a98dac644e254f97c20fb6d825f7e202fab213 --- /dev/null +++ b/k32-sae-mlp-768-seed2/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46655426fcb2ea695eaf989d482d5a506c403f0d90388a363978fd1d7f2f81ce +size 1268 diff --git a/k32-sae-mlp-768-seed2/optimizer.pt b/k32-sae-mlp-768-seed2/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..892fc3a95f5a3163ef8db9406c30a1214048217e --- /dev/null +++ b/k32-sae-mlp-768-seed2/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a673cb8bbcd01503613781201199d36d85ba225cd6e906a306b40322f91124a +size 28947578 diff --git a/k32-sae-mlp-768-seed2/state.pt b/k32-sae-mlp-768-seed2/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..29cc2a0245360233f500d7921b350f1fe020bcd5 --- /dev/null +++ b/k32-sae-mlp-768-seed2/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1022dacdf7289d8987798ebcc493bc45057e7991f602d9929fa77433d773a6eb +size 77138 diff --git a/k32-sae-mlp-768/config.json b/k32-sae-mlp-768/config.json new file mode 100644 index 0000000000000000000000000000000000000000..cfb66d3153b479345a35a4058e90e3821bd81f4a --- /dev/null +++ b/k32-sae-mlp-768/config.json @@ -0,0 +1 @@ +{"sae": {"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false}, "batch_size": 8, "grad_acc_steps": 1, "micro_acc_steps": 1, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0.mlp", "layers.1.mlp", "layers.2.mlp", "layers.3.mlp", "layers.4.mlp", "layers.5.mlp", "layers.6.mlp", "layers.7.mlp", "layers.8.mlp", "layers.9.mlp", "layers.10.mlp", "layers.11.mlp"], "layers": [], "layer_stride": 1, "transcode": false, "distribute_modules": false, "save_every": 1000, "log_to_wandb": true, "run_name": "k32-sae-mlp-768", "wandb_log_frequency": 1, "model": "EleutherAI/pythia-160m", "dataset": "/mnt/ssd-1/pile_preshuffled/standard/document.bin", "split": "train", "ctx_len": 2049, "hf_token": null, "revision": null, "load_in_8bit": false, "max_examples": 4000000, "resume": false, "finetune": null, "seed": 42, "data_preprocessing_num_proc": 48} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.0.mlp/cfg.json b/k32-sae-mlp-768/layers.0.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.0.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.0.mlp/sae.safetensors b/k32-sae-mlp-768/layers.0.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d23edd5546ab69bd914dd7f3e26dbd4b3bf13ea5 --- /dev/null +++ b/k32-sae-mlp-768/layers.0.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7614a43b3ae994004a221c82b9e60d61fb71848f9321ec55ac2ab74a5488488 +size 4725048 diff --git a/k32-sae-mlp-768/layers.1.mlp/cfg.json b/k32-sae-mlp-768/layers.1.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.1.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.1.mlp/sae.safetensors b/k32-sae-mlp-768/layers.1.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..39f392aa878632938673389199185f4930709312 --- /dev/null +++ b/k32-sae-mlp-768/layers.1.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2515f319716d0dfec91bc2182e7b529df72783c4f2ead97dbd6d80fd3935fa4b +size 4725048 diff --git a/k32-sae-mlp-768/layers.10.mlp/cfg.json b/k32-sae-mlp-768/layers.10.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.10.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.10.mlp/sae.safetensors b/k32-sae-mlp-768/layers.10.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..40f5ffe74c4f9d5010307fbcd0191e06e7342cf7 --- /dev/null +++ b/k32-sae-mlp-768/layers.10.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f637f55e7ec69a6987d519c9bd33b7952006029eb150a42323ccfeb5bd37f1a5 +size 4725048 diff --git a/k32-sae-mlp-768/layers.11.mlp/cfg.json b/k32-sae-mlp-768/layers.11.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.11.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.11.mlp/sae.safetensors b/k32-sae-mlp-768/layers.11.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a69cf2b2448536201b12ed3022529848b75b83de --- /dev/null +++ b/k32-sae-mlp-768/layers.11.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c1d5b7cdbfd9461063b46fc6f3c857e0855789f298eb6d4d559a60aba36c1ce4 +size 4725048 diff --git a/k32-sae-mlp-768/layers.2.mlp/cfg.json b/k32-sae-mlp-768/layers.2.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.2.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.2.mlp/sae.safetensors b/k32-sae-mlp-768/layers.2.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..af5d60d40dd22852badbc7acf0bf6254b8efa23c --- /dev/null +++ b/k32-sae-mlp-768/layers.2.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3307bde486a508327781dbd374c2871c9c20fc76fbe8a00b7ad7c46285165170 +size 4725048 diff --git a/k32-sae-mlp-768/layers.3.mlp/cfg.json b/k32-sae-mlp-768/layers.3.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.3.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.3.mlp/sae.safetensors b/k32-sae-mlp-768/layers.3.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..de25d4dc4397ae98d4991dd45e120527da18e481 --- /dev/null +++ b/k32-sae-mlp-768/layers.3.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0911261b3649db5831e0ad207960bc5d26de87ef2856ccf1b9bc1636dd26d53f +size 4725048 diff --git a/k32-sae-mlp-768/layers.4.mlp/cfg.json b/k32-sae-mlp-768/layers.4.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.4.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.4.mlp/sae.safetensors b/k32-sae-mlp-768/layers.4.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4eae283c0c4ef507c8c00e0f87b794757fd01eb7 --- /dev/null +++ b/k32-sae-mlp-768/layers.4.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23f882c21ee35e0730a0aaa0d06b394f205f4aa382d570978a57697bab7bab64 +size 4725048 diff --git a/k32-sae-mlp-768/layers.5.mlp/cfg.json b/k32-sae-mlp-768/layers.5.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.5.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.5.mlp/sae.safetensors b/k32-sae-mlp-768/layers.5.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..3aa94e99ffa2046a103b84810c7f08486b1c8aa3 --- /dev/null +++ b/k32-sae-mlp-768/layers.5.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30d841d56cd246d8ae673f6748069db2e65e26ad774b385ff6a4a1152ca1a248 +size 4725048 diff --git a/k32-sae-mlp-768/layers.6.mlp/cfg.json b/k32-sae-mlp-768/layers.6.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.6.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.6.mlp/sae.safetensors b/k32-sae-mlp-768/layers.6.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..61ebfa6391cd43b91d95aea0a46278e69f176cb4 --- /dev/null +++ b/k32-sae-mlp-768/layers.6.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1cc88ed824c27a8d5255f3f9193516b802d7da6d3aef8725d4a055b9dcc9a6dd +size 4725048 diff --git a/k32-sae-mlp-768/layers.7.mlp/cfg.json b/k32-sae-mlp-768/layers.7.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.7.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.7.mlp/sae.safetensors b/k32-sae-mlp-768/layers.7.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a25bcd5b0ecc173f13c9b1ca73cdc77190693295 --- /dev/null +++ b/k32-sae-mlp-768/layers.7.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9109097c677fa6b978f17e16b622adfb78328ac8d1f847f58e56b84ef5f412ee +size 4725048 diff --git a/k32-sae-mlp-768/layers.8.mlp/cfg.json b/k32-sae-mlp-768/layers.8.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.8.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.8.mlp/sae.safetensors b/k32-sae-mlp-768/layers.8.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..eafbdfe97ca17c7f47bf339f3176056b12dc8b3c --- /dev/null +++ b/k32-sae-mlp-768/layers.8.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10d817e80557517da5413966d76e5664efbdea933ed9f2f839fb02a1d2397a58 +size 4725048 diff --git a/k32-sae-mlp-768/layers.9.mlp/cfg.json b/k32-sae-mlp-768/layers.9.mlp/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..d8daf5ea2d5574c0c44cc6b0278c650aa01bcc8a --- /dev/null +++ b/k32-sae-mlp-768/layers.9.mlp/cfg.json @@ -0,0 +1 @@ +{"expansion_factor": 32, "normalize_decoder": true, "num_latents": 768, "k": 32, "multi_topk": false, "skip_connection": false, "d_in": 768} \ No newline at end of file diff --git a/k32-sae-mlp-768/layers.9.mlp/sae.safetensors b/k32-sae-mlp-768/layers.9.mlp/sae.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4f0f5a9e5cbbd4c49372900897b81c932e1f3ef8 --- /dev/null +++ b/k32-sae-mlp-768/layers.9.mlp/sae.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fee9934b042aa6269ae3df0a2b99ed29c7ddc50faf2df0ef898d3e6bcfec74f +size 4725048 diff --git a/k32-sae-mlp-768/lr_scheduler.pt b/k32-sae-mlp-768/lr_scheduler.pt new file mode 100644 index 0000000000000000000000000000000000000000..18a98dac644e254f97c20fb6d825f7e202fab213 --- /dev/null +++ b/k32-sae-mlp-768/lr_scheduler.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46655426fcb2ea695eaf989d482d5a506c403f0d90388a363978fd1d7f2f81ce +size 1268 diff --git a/k32-sae-mlp-768/optimizer.pt b/k32-sae-mlp-768/optimizer.pt new file mode 100644 index 0000000000000000000000000000000000000000..d932e1e4ae67014371374209b8ce069b8214ba4a --- /dev/null +++ b/k32-sae-mlp-768/optimizer.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47ade93285e4ee2626dff3540fa65416a2e57e880ccf0170988c571acec7100a +size 28947578 diff --git a/k32-sae-mlp-768/state.pt b/k32-sae-mlp-768/state.pt new file mode 100644 index 0000000000000000000000000000000000000000..0f5ca5652e3567d1f06a9ce6ab8dc8bbab140d67 --- /dev/null +++ b/k32-sae-mlp-768/state.pt @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89a0aeeb13376ffa3adb4ad040c1e02ac7ecb7591b9d0b95c4ad8d8ce6667e6d +size 77138