diff --git a/.gitattributes b/.gitattributes index a6344aac8c09253b3b630fb776ae94478aa0275b..e993dca7b2554ac0d56ad8267b153248b4253315 100644 --- a/.gitattributes +++ b/.gitattributes @@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text *.zip filter=lfs diff=lfs merge=lfs -text *.zst filter=lfs diff=lfs merge=lfs -text *tfevents* filter=lfs diff=lfs merge=lfs -text +benchmark_stats.png filter=lfs diff=lfs merge=lfs -text diff --git a/benchmark_stats.csv b/benchmark_stats.csv new file mode 100644 index 0000000000000000000000000000000000000000..aa2984f0b86e5da129b8b8d872140894666550b2 --- /dev/null +++ b/benchmark_stats.csv @@ -0,0 +1,13 @@ +,version,d_sae,layer,kl_div_with_sae,kl_div_with_ablation,ce_loss_with_sae,ce_loss_without_sae,ce_loss_with_ablation,kl_div_score,ce_loss_score,l2_norm_in,l2_norm_out,l2_ratio,l0,l1,explained_variance,mse,total_tokens_evaluated,filepath +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_0/metrics.json,5,32,0,0.004214919172227383,2.121527671813965,3.6037631034851074,3.599064588546753,5.74860143661499,0.9980132622222063,0.9978141733450269,32.01313781738281,31.89154624938965,0.9962403178215027,31.99397850036621,42.146968841552734,0.9667115211486816,8.074386596679688,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_0/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_1/metrics.json,5,32,1,0.001881452277302742,0.024065840989351273,3.6013145446777344,3.599064588546753,3.6206326484680176,0.9218206304057585,0.895681107192996,9.714648246765137,9.157854080200195,0.9379116892814636,32.0,82.86055755615234,0.8680867552757263,9.54197883605957,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_1/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_2/metrics.json,5,32,2,0.002341545419767499,0.031004613265395164,3.6009159088134766,3.599064588546753,3.626660108566284,0.9244775156611632,0.9329122891899364,8.641822814941406,8.045538902282715,0.9296190738677979,32.0,82.86361694335938,0.8532325029373169,9.74060344696045,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_2/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_3/metrics.json,5,32,3,0.0028422873001545668,0.025133918970823288,3.602360486984253,3.599064588546753,3.6286609172821045,0.8869142809183862,0.8886382677063865,8.571012496948242,7.753783226013184,0.9047597646713257,32.0,81.56338500976562,0.8156890869140625,13.545696258544922,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_3/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_4/metrics.json,5,32,4,0.003790093120187521,0.026722650974988937,3.603180408477783,3.599064588546753,3.6321334838867188,0.8581692690693428,0.8755380278440674,9.123016357421875,7.9935712814331055,0.8777990341186523,32.0,79.12753295898438,0.7723549604415894,19.59968376159668,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_4/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_5/metrics.json,5,32,5,0.004055157769471407,0.031378373503685,3.602062940597534,3.599064588546753,3.6277596950531006,0.8707658391218022,0.8955099870384526,10.034396171569824,8.880256652832031,0.8862426280975342,32.0,77.73406219482422,0.7825504541397095,24.739639282226562,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_5/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_6/metrics.json,5,32,6,0.005056389141827822,0.03278880566358566,3.604351758956909,3.599064588546753,3.634286403656006,0.8457891637253715,0.8498893258693165,11.67806625366211,10.210134506225586,0.877007007598877,32.0,74.85708618164062,0.7534131407737732,35.03990936279297,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_6/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_7/metrics.json,5,32,7,0.004875506274402142,0.03466065973043442,3.6051602363586426,3.599064588546753,3.634833812713623,0.8593360220976661,0.8295840082118552,13.65020751953125,12.291288375854492,0.9022888541221619,32.0,71.10636901855469,0.7838281393051147,41.60253143310547,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_7/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_8/metrics.json,5,32,8,0.00555825512856245,0.02938206121325493,3.6046571731567383,3.599064588546753,3.6258018016815186,0.8108282775595406,0.7908314310172635,16.137948989868164,14.443827629089355,0.8963184356689453,32.0,71.65403747558594,0.759009063243866,57.211456298828125,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_8/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_9/metrics.json,5,32,9,0.004498911090195179,0.028918448835611343,3.6015830039978027,3.599064588546753,3.636500835418701,0.8444276483925708,0.9327278864341259,20.912498474121094,19.139347076416016,0.9177886843681335,32.0,65.82906341552734,0.7807904481887817,77.84382629394531,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_9/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_10/metrics.json,5,32,10,0.003998876549303532,0.02475181221961975,3.602677822113037,3.599064588546753,3.6404881477355957,0.8384410598374779,0.912773464254675,31.82137680053711,30.121129989624023,0.9457573890686035,32.0,55.563880920410156,0.819293737411499,125.34260559082031,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_10/metrics.json +OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_11/metrics.json,5,32,11,0.0037718701642006636,0.10687470436096191,3.6011340618133545,3.599064588546753,3.730869770050049,0.9647075499599846,0.9842989991516394,280.86444091796875,280.543212890625,0.9986675977706909,31.6875,17.145309448242188,0.9678490161895752,180.27102661132812,6144.0,OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_11/metrics.json diff --git a/benchmark_stats.html b/benchmark_stats.html new file mode 100644 index 0000000000000000000000000000000000000000..24cb7eb5e92ad56a9c9426a620b83e008cf00000 --- /dev/null +++ b/benchmark_stats.html @@ -0,0 +1,618 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
 versiond_saelayerkl_div_with_saekl_div_with_ablationce_loss_with_saece_loss_without_saece_loss_with_ablationkl_div_scorece_loss_scorel2_norm_inl2_norm_outl2_ratiol0l1explained_variancemsetotal_tokens_evaluatedfilepath
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_0/metrics.json53200.0042152.1215283.6037633.5990655.7486010.9980130.99781432.01313831.8915460.99624031.99397942.1469690.9667128.0743876144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_0/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_1/metrics.json53210.0018810.0240663.6013153.5990653.6206330.9218210.8956819.7146489.1578540.93791232.00000082.8605580.8680879.5419796144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_1/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_2/metrics.json53220.0023420.0310053.6009163.5990653.6266600.9244780.9329128.6418238.0455390.92961932.00000082.8636170.8532339.7406036144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_2/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_3/metrics.json53230.0028420.0251343.6023603.5990653.6286610.8869140.8886388.5710127.7537830.90476032.00000081.5633850.81568913.5456966144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_3/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_4/metrics.json53240.0037900.0267233.6031803.5990653.6321330.8581690.8755389.1230167.9935710.87779932.00000079.1275330.77235519.5996846144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_4/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_5/metrics.json53250.0040550.0313783.6020633.5990653.6277600.8707660.89551010.0343968.8802570.88624332.00000077.7340620.78255024.7396396144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_5/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_6/metrics.json53260.0050560.0327893.6043523.5990653.6342860.8457890.84988911.67806610.2101350.87700732.00000074.8570860.75341335.0399096144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_6/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_7/metrics.json53270.0048760.0346613.6051603.5990653.6348340.8593360.82958413.65020812.2912880.90228932.00000071.1063690.78382841.6025316144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_7/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_8/metrics.json53280.0055580.0293823.6046573.5990653.6258020.8108280.79083116.13794914.4438280.89631832.00000071.6540370.75900957.2114566144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_8/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_9/metrics.json53290.0044990.0289183.6015833.5990653.6365010.8444280.93272820.91249819.1393470.91778932.00000065.8290630.78079077.8438266144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_9/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_10/metrics.json532100.0039990.0247523.6026783.5990653.6404880.8384410.91277331.82137730.1211300.94575732.00000055.5638810.819294125.3426066144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_10/metrics.json
OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_11/metrics.json532110.0037720.1068753.6011343.5990653.7308700.9647080.984299280.864441280.5432130.99866831.68750017.1453090.967849180.2710276144.000000OAI_GPT2Small_v5_32k_resid_delta_attn/v5_32k_layer_11/metrics.json
diff --git a/benchmark_stats.png b/benchmark_stats.png new file mode 100644 index 0000000000000000000000000000000000000000..b6539415163091f8ff57ac1d1225cc0229514dba --- /dev/null +++ b/benchmark_stats.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:baa1f02359b614167ee31073a100b297687e1f6810662ad15bf1de9bf8f9fc5a +size 4559124 diff --git a/v5_32k_layer_0/cfg.json b/v5_32k_layer_0/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..33829c583dfc50e64f2935f5e2a2c9f3169a36ba --- /dev/null +++ b/v5_32k_layer_0/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.0.hook_attn_out", "hook_layer": 0, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_0/metrics.json b/v5_32k_layer_0/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..0b4d7eaa7629a5b8ba06a22360b6cf85be21d8b7 --- /dev/null +++ b/v5_32k_layer_0/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.004214919172227383, "metrics/kl_div_with_ablation": 2.121527671813965, "metrics/ce_loss_with_sae": 3.6037631034851074, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 5.74860143661499, "metrics/kl_div_score": 0.9980132622222063, "metrics/ce_loss_score": 0.9978141733450269, "metrics/l2_norm_in": 32.01313781738281, "metrics/l2_norm_out": 31.89154624938965, "metrics/l2_ratio": 0.9962403178215027, "metrics/l0": 31.99397850036621, "metrics/l1": 42.146968841552734, "metrics/explained_variance": 0.9667115211486816, "metrics/mse": 8.074386596679688, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_0/sae_weights.safetensors b/v5_32k_layer_0/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..6ac385643297adee68c66859a0f965c749044616 --- /dev/null +++ b/v5_32k_layer_0/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8209cd0dacf3c9638052b5dcb6eed059f1f9cb8da440faab92758b2befdd21eb +size 201461056 diff --git a/v5_32k_layer_0/sparsity.safetensors b/v5_32k_layer_0/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..003c21522f817ce8694b39e55d1207dfea2a0e28 --- /dev/null +++ b/v5_32k_layer_0/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8265db43d0be97981e9be43b9c6bd468c27a14e3fd1b60d4290779d5e3929528 +size 131152 diff --git a/v5_32k_layer_1/cfg.json b/v5_32k_layer_1/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..a1c35416f5ca1fc26e959d7d08f682481fbe6fcd --- /dev/null +++ b/v5_32k_layer_1/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.1.hook_attn_out", "hook_layer": 1, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_1/metrics.json b/v5_32k_layer_1/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..d907f51875878e9421d814ffa2e5ba694e975ae9 --- /dev/null +++ b/v5_32k_layer_1/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.001881452277302742, "metrics/kl_div_with_ablation": 0.024065840989351273, "metrics/ce_loss_with_sae": 3.6013145446777344, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6206326484680176, "metrics/kl_div_score": 0.9218206304057585, "metrics/ce_loss_score": 0.895681107192996, "metrics/l2_norm_in": 9.714648246765137, "metrics/l2_norm_out": 9.157854080200195, "metrics/l2_ratio": 0.9379116892814636, "metrics/l0": 32.0, "metrics/l1": 82.86055755615234, "metrics/explained_variance": 0.8680867552757263, "metrics/mse": 9.54197883605957, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_1/sae_weights.safetensors b/v5_32k_layer_1/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1d9845dafe8096ce14da9822c8dff70de9d9e28a --- /dev/null +++ b/v5_32k_layer_1/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:227271d3fb262bedb7f2fa188f28e3a05c4a623817f00985681b3b42d66bbb76 +size 201461056 diff --git a/v5_32k_layer_1/sparsity.safetensors b/v5_32k_layer_1/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1a1ec5ef9f652ec40f5e84bf4326ea52d2299be9 --- /dev/null +++ b/v5_32k_layer_1/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f43a8be433ddcb0d02d93d37ec7edf47f4fa8263151f8a342d8a153e658e9fb1 +size 131152 diff --git a/v5_32k_layer_10/cfg.json b/v5_32k_layer_10/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..9c562e5812095efc6236500bd5c8137decb232e7 --- /dev/null +++ b/v5_32k_layer_10/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.10.hook_attn_out", "hook_layer": 10, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_10/metrics.json b/v5_32k_layer_10/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..e296c6ae635657b7c209644f389288f32303fbea --- /dev/null +++ b/v5_32k_layer_10/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.003998876549303532, "metrics/kl_div_with_ablation": 0.02475181221961975, "metrics/ce_loss_with_sae": 3.602677822113037, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6404881477355957, "metrics/kl_div_score": 0.8384410598374779, "metrics/ce_loss_score": 0.912773464254675, "metrics/l2_norm_in": 31.82137680053711, "metrics/l2_norm_out": 30.121129989624023, "metrics/l2_ratio": 0.9457573890686035, "metrics/l0": 32.0, "metrics/l1": 55.563880920410156, "metrics/explained_variance": 0.819293737411499, "metrics/mse": 125.34260559082031, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_10/sae_weights.safetensors b/v5_32k_layer_10/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..64d63a1d1c1cf7d1eb8abbd64dcfc166d3375476 --- /dev/null +++ b/v5_32k_layer_10/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2d384b39c05ee6a1951e730e3f7c8a6f4017bbb59b20b0d78f72edbe08984d1 +size 201461056 diff --git a/v5_32k_layer_10/sparsity.safetensors b/v5_32k_layer_10/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e736700f583a99a7f0e8867da9bcc37ed7f9ddf --- /dev/null +++ b/v5_32k_layer_10/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:534fd6f949997b868a8f558984957f79f1fd0902a3dd66da38e7148fb74529cf +size 131152 diff --git a/v5_32k_layer_11/cfg.json b/v5_32k_layer_11/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..3eed051fb6a34a1eb0b6e0825bfae4d60a58dc07 --- /dev/null +++ b/v5_32k_layer_11/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.11.hook_attn_out", "hook_layer": 11, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_11/metrics.json b/v5_32k_layer_11/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..8c0298983e22aa8e0ad81ec83d01e819de00c623 --- /dev/null +++ b/v5_32k_layer_11/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.0037718701642006636, "metrics/kl_div_with_ablation": 0.10687470436096191, "metrics/ce_loss_with_sae": 3.6011340618133545, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.730869770050049, "metrics/kl_div_score": 0.9647075499599846, "metrics/ce_loss_score": 0.9842989991516394, "metrics/l2_norm_in": 280.86444091796875, "metrics/l2_norm_out": 280.543212890625, "metrics/l2_ratio": 0.9986675977706909, "metrics/l0": 31.6875, "metrics/l1": 17.145309448242188, "metrics/explained_variance": 0.9678490161895752, "metrics/mse": 180.27102661132812, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_11/sae_weights.safetensors b/v5_32k_layer_11/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b9b5732ed82695a94ab9fec288404c8e7975922a --- /dev/null +++ b/v5_32k_layer_11/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9c4edd6ed3c82033d3be87829b9f7384a8c2a35cc1501c9a401dde0c60de470 +size 201461056 diff --git a/v5_32k_layer_11/sparsity.safetensors b/v5_32k_layer_11/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..1214e45ae0b933af94317fa3d121441b114307fe --- /dev/null +++ b/v5_32k_layer_11/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d0acf780230a6be893d1c7f32327b63d5176b59ceab2072a17a0e39ac8fcdc3 +size 131152 diff --git a/v5_32k_layer_2/cfg.json b/v5_32k_layer_2/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..89f520599723f20143549110f1f686d6c51ded10 --- /dev/null +++ b/v5_32k_layer_2/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.2.hook_attn_out", "hook_layer": 2, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_2/metrics.json b/v5_32k_layer_2/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..990c2fc2fc72e93afce7116fe82ab9405f91df08 --- /dev/null +++ b/v5_32k_layer_2/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.002341545419767499, "metrics/kl_div_with_ablation": 0.031004613265395164, "metrics/ce_loss_with_sae": 3.6009159088134766, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.626660108566284, "metrics/kl_div_score": 0.9244775156611632, "metrics/ce_loss_score": 0.9329122891899364, "metrics/l2_norm_in": 8.641822814941406, "metrics/l2_norm_out": 8.045538902282715, "metrics/l2_ratio": 0.9296190738677979, "metrics/l0": 32.0, "metrics/l1": 82.86361694335938, "metrics/explained_variance": 0.8532325029373169, "metrics/mse": 9.74060344696045, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_2/sae_weights.safetensors b/v5_32k_layer_2/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b8ee3009514ca8466b878fcee8f9617fff2d2ef9 --- /dev/null +++ b/v5_32k_layer_2/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db275872776ac34530b691887ad719316f0b00f92ad630f4fc7ec238cd50377d +size 201461056 diff --git a/v5_32k_layer_2/sparsity.safetensors b/v5_32k_layer_2/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..983dcf3ca8ff7d0c6bbce39b975f4c2292f6b8d6 --- /dev/null +++ b/v5_32k_layer_2/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5eb7d67b660dc6027b4543482780f14982d98f5bb0a4cdf3aa9738c86a7f5393 +size 131152 diff --git a/v5_32k_layer_3/cfg.json b/v5_32k_layer_3/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..6dd937e8324e64c702af2d800b82da24aeb8d730 --- /dev/null +++ b/v5_32k_layer_3/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.3.hook_attn_out", "hook_layer": 3, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_3/metrics.json b/v5_32k_layer_3/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..1b9fb85ae65df6770439e2604cade38874fc779c --- /dev/null +++ b/v5_32k_layer_3/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.0028422873001545668, "metrics/kl_div_with_ablation": 0.025133918970823288, "metrics/ce_loss_with_sae": 3.602360486984253, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6286609172821045, "metrics/kl_div_score": 0.8869142809183862, "metrics/ce_loss_score": 0.8886382677063865, "metrics/l2_norm_in": 8.571012496948242, "metrics/l2_norm_out": 7.753783226013184, "metrics/l2_ratio": 0.9047597646713257, "metrics/l0": 32.0, "metrics/l1": 81.56338500976562, "metrics/explained_variance": 0.8156890869140625, "metrics/mse": 13.545696258544922, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_3/sae_weights.safetensors b/v5_32k_layer_3/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4460edd9be6567da1d7f5ef52a56c84be9c76dce --- /dev/null +++ b/v5_32k_layer_3/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd02ed8e83ae92fd441d6eb62647c28f173ecd4a2c488140474489b6cc2a7b77 +size 201461056 diff --git a/v5_32k_layer_3/sparsity.safetensors b/v5_32k_layer_3/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4bf8548d51d982b43b8c501d64a0b080e7b47f14 --- /dev/null +++ b/v5_32k_layer_3/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b4478a72ad24dc5af709816db7e211db0b5526c75d0b028944c2d18f1098380 +size 131152 diff --git a/v5_32k_layer_4/cfg.json b/v5_32k_layer_4/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..45a9485a712278f3ae93a65d7daadcbe01956af5 --- /dev/null +++ b/v5_32k_layer_4/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.4.hook_attn_out", "hook_layer": 4, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_4/metrics.json b/v5_32k_layer_4/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..9564b00e3c840c81447096afdbe1e84c9507ea7e --- /dev/null +++ b/v5_32k_layer_4/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.003790093120187521, "metrics/kl_div_with_ablation": 0.026722650974988937, "metrics/ce_loss_with_sae": 3.603180408477783, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6321334838867188, "metrics/kl_div_score": 0.8581692690693428, "metrics/ce_loss_score": 0.8755380278440674, "metrics/l2_norm_in": 9.123016357421875, "metrics/l2_norm_out": 7.9935712814331055, "metrics/l2_ratio": 0.8777990341186523, "metrics/l0": 32.0, "metrics/l1": 79.12753295898438, "metrics/explained_variance": 0.7723549604415894, "metrics/mse": 19.59968376159668, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_4/sae_weights.safetensors b/v5_32k_layer_4/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fea7f3fc9304b709cbafedafbf687363b48f84d0 --- /dev/null +++ b/v5_32k_layer_4/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ff34559202308047f7e25ef2002a023dfdbdaf12002618b1f3f4f74f2bf4307 +size 201461056 diff --git a/v5_32k_layer_4/sparsity.safetensors b/v5_32k_layer_4/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c1c762ed84d480920ade653fe1fd33d58f712816 --- /dev/null +++ b/v5_32k_layer_4/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76e2f66395864ba52cb8da7a5d6f13e14d8e3cc44f6b21439a83ff9eb6c0ac6f +size 131152 diff --git a/v5_32k_layer_5/cfg.json b/v5_32k_layer_5/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..db0986e7684d6372854afa393e22f2a716cf23cd --- /dev/null +++ b/v5_32k_layer_5/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.5.hook_attn_out", "hook_layer": 5, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_5/metrics.json b/v5_32k_layer_5/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..ce1d9fa62362859a3bb032490eb0014ccc7b5fe0 --- /dev/null +++ b/v5_32k_layer_5/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.004055157769471407, "metrics/kl_div_with_ablation": 0.031378373503685, "metrics/ce_loss_with_sae": 3.602062940597534, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6277596950531006, "metrics/kl_div_score": 0.8707658391218022, "metrics/ce_loss_score": 0.8955099870384526, "metrics/l2_norm_in": 10.034396171569824, "metrics/l2_norm_out": 8.880256652832031, "metrics/l2_ratio": 0.8862426280975342, "metrics/l0": 32.0, "metrics/l1": 77.73406219482422, "metrics/explained_variance": 0.7825504541397095, "metrics/mse": 24.739639282226562, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_5/sae_weights.safetensors b/v5_32k_layer_5/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ed66f631701f6d0d6ec9d12b1028418ea789e9b5 --- /dev/null +++ b/v5_32k_layer_5/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47166d73eccc163c3738446b07623ea7e4ed85af51c59c19ebbc71385b2e5e76 +size 201461056 diff --git a/v5_32k_layer_5/sparsity.safetensors b/v5_32k_layer_5/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f15f8b0e847e8e81a1e99734c5d5a17da761cfbf --- /dev/null +++ b/v5_32k_layer_5/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87a88d9c69a1332d9e037690e0fdcd20c11854d02406243b39afca769f29b661 +size 131152 diff --git a/v5_32k_layer_6/cfg.json b/v5_32k_layer_6/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..f143b3f9e31b34db4447a40dd0c86d1ebb402ccf --- /dev/null +++ b/v5_32k_layer_6/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.6.hook_attn_out", "hook_layer": 6, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_6/metrics.json b/v5_32k_layer_6/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..b2019c18b538393863d291863ba95880cbc94f5d --- /dev/null +++ b/v5_32k_layer_6/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.005056389141827822, "metrics/kl_div_with_ablation": 0.03278880566358566, "metrics/ce_loss_with_sae": 3.604351758956909, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.634286403656006, "metrics/kl_div_score": 0.8457891637253715, "metrics/ce_loss_score": 0.8498893258693165, "metrics/l2_norm_in": 11.67806625366211, "metrics/l2_norm_out": 10.210134506225586, "metrics/l2_ratio": 0.877007007598877, "metrics/l0": 32.0, "metrics/l1": 74.85708618164062, "metrics/explained_variance": 0.7534131407737732, "metrics/mse": 35.03990936279297, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_6/sae_weights.safetensors b/v5_32k_layer_6/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..37aaf5fcc613c4beb4809fbf1a1d8116185b6697 --- /dev/null +++ b/v5_32k_layer_6/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f68fbfac96d5b88f59bdc8dc57049b95f048d88bb534b8f7b0f9fb4d640ecef +size 201461056 diff --git a/v5_32k_layer_6/sparsity.safetensors b/v5_32k_layer_6/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ecdfc98f4cf406aff96728a1692e4c2e388603be --- /dev/null +++ b/v5_32k_layer_6/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f51bb77a2d34541c0fffe40c4311b9be1be3a15d8b229a2959bfc488eca0c27 +size 131152 diff --git a/v5_32k_layer_7/cfg.json b/v5_32k_layer_7/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..c4187ab6de53af97ce6dfdf810fd0d1d234b774d --- /dev/null +++ b/v5_32k_layer_7/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.7.hook_attn_out", "hook_layer": 7, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_7/metrics.json b/v5_32k_layer_7/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..e8a9e7c2509b44e90e93026532a4d8359e68cd76 --- /dev/null +++ b/v5_32k_layer_7/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.004875506274402142, "metrics/kl_div_with_ablation": 0.03466065973043442, "metrics/ce_loss_with_sae": 3.6051602363586426, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.634833812713623, "metrics/kl_div_score": 0.8593360220976661, "metrics/ce_loss_score": 0.8295840082118552, "metrics/l2_norm_in": 13.65020751953125, "metrics/l2_norm_out": 12.291288375854492, "metrics/l2_ratio": 0.9022888541221619, "metrics/l0": 32.0, "metrics/l1": 71.10636901855469, "metrics/explained_variance": 0.7838281393051147, "metrics/mse": 41.60253143310547, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_7/sae_weights.safetensors b/v5_32k_layer_7/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9e937a587b1dff5f8d41168460963ec6325a319e --- /dev/null +++ b/v5_32k_layer_7/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7de864c425755808d51cd49f65164f0fba524923ae7d0588b1c4c808ffc325fc +size 201461056 diff --git a/v5_32k_layer_7/sparsity.safetensors b/v5_32k_layer_7/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..191d31a4a590479aae7ef4080a363cde7106a69b --- /dev/null +++ b/v5_32k_layer_7/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:900dca3640c1fe08bb1e87c3b8d2c51d7bd507aa752543f56926ebd157cb65c1 +size 131152 diff --git a/v5_32k_layer_8/cfg.json b/v5_32k_layer_8/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..e6e18e64eeaac74050bbefbf08d9e558e2ec07d4 --- /dev/null +++ b/v5_32k_layer_8/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.8.hook_attn_out", "hook_layer": 8, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_8/metrics.json b/v5_32k_layer_8/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..ecc41ebff815fcaf1cb5b725372b16d333963cfe --- /dev/null +++ b/v5_32k_layer_8/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.00555825512856245, "metrics/kl_div_with_ablation": 0.02938206121325493, "metrics/ce_loss_with_sae": 3.6046571731567383, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6258018016815186, "metrics/kl_div_score": 0.8108282775595406, "metrics/ce_loss_score": 0.7908314310172635, "metrics/l2_norm_in": 16.137948989868164, "metrics/l2_norm_out": 14.443827629089355, "metrics/l2_ratio": 0.8963184356689453, "metrics/l0": 32.0, "metrics/l1": 71.65403747558594, "metrics/explained_variance": 0.759009063243866, "metrics/mse": 57.211456298828125, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_8/sae_weights.safetensors b/v5_32k_layer_8/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..ff4a407d9219230c3bb010242d609721eacbb78c --- /dev/null +++ b/v5_32k_layer_8/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d45d5761057c9af25fa6b664409ecdd837d172fde59746647d50ae93f15b481d +size 201461056 diff --git a/v5_32k_layer_8/sparsity.safetensors b/v5_32k_layer_8/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..86bfde3478d45bb9ccc8bf16f1fe6ba590a88d4b --- /dev/null +++ b/v5_32k_layer_8/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1fcff9d1cc927cecfe615a28634cd3a61bca7af5092a0fe05194f8e05cf843f6 +size 131152 diff --git a/v5_32k_layer_9/cfg.json b/v5_32k_layer_9/cfg.json new file mode 100644 index 0000000000000000000000000000000000000000..8d73a99e13750cb1a69cd68f47286387d80273f8 --- /dev/null +++ b/v5_32k_layer_9/cfg.json @@ -0,0 +1 @@ +{"architecture": "standard", "d_in": 768, "d_sae": 32768, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.9.hook_attn_out", "hook_layer": 9, "hook_head_index": null, "activation_fn_str": "topk", "activation_fn_kwargs": {"k": 32}, "apply_b_dec_to_input": true, "finetuning_scaling_factor": false, "sae_lens_training_version": null, "prepend_bos": false, "dataset_path": "Skylion007/openwebtext", "dataset_trust_remote_code": true, "context_size": 64, "normalize_activations": "layer_norm"} \ No newline at end of file diff --git a/v5_32k_layer_9/metrics.json b/v5_32k_layer_9/metrics.json new file mode 100644 index 0000000000000000000000000000000000000000..6167600c1399f762e6023312156d3fca90df506a --- /dev/null +++ b/v5_32k_layer_9/metrics.json @@ -0,0 +1 @@ +{"metrics/kl_div_with_sae": 0.004498911090195179, "metrics/kl_div_with_ablation": 0.028918448835611343, "metrics/ce_loss_with_sae": 3.6015830039978027, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.636500835418701, "metrics/kl_div_score": 0.8444276483925708, "metrics/ce_loss_score": 0.9327278864341259, "metrics/l2_norm_in": 20.912498474121094, "metrics/l2_norm_out": 19.139347076416016, "metrics/l2_ratio": 0.9177886843681335, "metrics/l0": 32.0, "metrics/l1": 65.82906341552734, "metrics/explained_variance": 0.7807904481887817, "metrics/mse": 77.84382629394531, "metrics/total_tokens_evaluated": 6144} \ No newline at end of file diff --git a/v5_32k_layer_9/sae_weights.safetensors b/v5_32k_layer_9/sae_weights.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b5c4400dc911d4e641c6119e1a541b45b9d89613 --- /dev/null +++ b/v5_32k_layer_9/sae_weights.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b687be7b963de35bf7de8089a3ce5e5af6f62d5b36014a32031190c4c87043da +size 201461056 diff --git a/v5_32k_layer_9/sparsity.safetensors b/v5_32k_layer_9/sparsity.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..4914eb9086dd4f0026429189d057b216af1af0a7 --- /dev/null +++ b/v5_32k_layer_9/sparsity.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee61addec2ff5590f4c0d2642674e2f39f678966224728b11fab4ee92db7e578 +size 131152