Upload folder using huggingface_hub

.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

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_128k_resid_delta_mlp/v5_128k_layer_0/metrics.json,5,128,0,0.0028826396446675062,3.094083309173584,3.6018943786621094,3.599064588546753,6.694648742675781,0.999068338064421,0.9990858623204987,29.933448791503906,29.71919822692871,0.9930059909820557,31.99934959411621,61.198638916015625,0.9775193929672241,13.276110649108887,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_0/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_1/metrics.json,5,128,1,0.004241525661200285,0.05105271190404892,3.599820852279663,3.599064588546753,3.6525371074676514,0.9169187002411934,0.9858569645086499,18.973735809326172,18.205364227294922,0.9327648878097534,32.0,84.0601806640625,0.9169172048568726,17.682374954223633,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_1/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_2/metrics.json,5,128,2,0.006751821842044592,0.05874736234545708,3.602034091949463,3.599064588546753,3.6459126472473145,0.8850702129852012,0.9366141632102598,49.106536865234375,47.9766845703125,0.912467360496521,31.984375,82.67613983154297,0.981090247631073,28.213790893554688,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_2/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_3/metrics.json,5,128,3,0.008531898260116577,0.07059153914451599,3.6077229976654053,3.599064588546753,3.658677577972412,0.8791370982484182,0.8547563341132242,16.98731803894043,15.537837028503418,0.8995180130004883,31.90511131286621,81.88644409179688,0.8274940252304077,38.814979553222656,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_3/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_4/metrics.json,5,128,4,0.010455957613885403,0.06332532316446304,3.6118924617767334,3.599064588546753,3.6600804328918457,0.8348850492759414,0.7897616042575971,17.251985549926758,15.434064865112305,0.8794459700584412,31.984375,78.37165832519531,0.780777633190155,50.81820297241211,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_4/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_5/metrics.json,5,128,5,0.011617729440331459,0.06850520521402359,3.6092593669891357,3.599064588546753,3.6693859100341797,0.8304109971784566,0.8550257841186103,18.888967514038086,16.669851303100586,0.8737667798995972,31.98372459411621,77.13916015625,0.7653716802597046,71.50291442871094,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_5/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_6/metrics.json,5,128,6,0.014023388735949993,0.07569437474012375,3.6142406463623047,3.599064588546753,3.67651629447937,0.814736712152052,0.8040577979166154,21.466564178466797,18.890602111816406,0.8757946491241455,31.99837303161621,74.61764526367188,0.754051685333252,97.22125244140625,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_6/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_7/metrics.json,5,128,7,0.014005273580551147,0.08048636466264725,3.6098315715789795,3.599064588546753,3.6727118492126465,0.8259919721899077,0.8538033467249813,25.444438934326172,22.559194564819336,0.8843803405761719,31.99788475036621,72.55130004882812,0.7648060321807861,130.78082275390625,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_7/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_8/metrics.json,5,128,8,0.0149507075548172,0.08732372522354126,3.6158981323242188,3.599064588546753,3.6803367137908936,0.828789856175459,0.7928743252757569,30.250225067138672,26.934555053710938,0.8884322047233582,31.99690818786621,72.55912780761719,0.7680301070213318,183.63497924804688,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_8/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_9/metrics.json,5,128,9,0.01591905765235424,0.09758877754211426,3.615844488143921,3.599064588546753,3.6962451934814453,0.8368761444369522,0.827332834484366,40.192413330078125,36.57688903808594,0.9057919383049011,32.0,68.9762954711914,0.7820959091186523,267.2160949707031,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_9/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_10/metrics.json,5,128,10,0.019077030941843987,0.12674781680107117,3.616905927658081,3.599064588546753,3.708984136581421,0.8494882876619082,0.8376872955691096,81.75682830810547,78.83488464355469,0.9609333872795105,32.0,48.323631286621094,0.819786787033081,443.10888671875,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_10/metrics.json
+OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_11/metrics.json,5,128,11,0.024071572348475456,0.17384076118469238,3.620229721069336,3.599064588546753,3.783318281173706,0.8615309080308198,0.8851304838409143,92.90629577636719,88.29003143310547,0.9321191310882568,32.0,71.25521850585938,0.8613338470458984,635.45947265625,6144.0,OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_11/metrics.json

benchmark_stats.html

@@ -0,0 +1,674 @@
+<style type="text/css">
+#T_a780d_row0_col2, #T_a780d_row0_col3, #T_a780d_row0_col6, #T_a780d_row0_col16, #T_a780d_row0_col17, #T_a780d_row1_col4, #T_a780d_row1_col5, #T_a780d_row1_col6, #T_a780d_row1_col7, #T_a780d_row1_col17, #T_a780d_row2_col4, #T_a780d_row2_col6, #T_a780d_row2_col7, #T_a780d_row2_col17, #T_a780d_row3_col6, #T_a780d_row3_col10, #T_a780d_row3_col11, #T_a780d_row3_col13, #T_a780d_row3_col17, #T_a780d_row4_col6, #T_a780d_row4_col9, #T_a780d_row4_col10, #T_a780d_row4_col11, #T_a780d_row4_col17, #T_a780d_row5_col6, #T_a780d_row5_col12, #T_a780d_row5_col17, #T_a780d_row6_col6, #T_a780d_row6_col8, #T_a780d_row6_col15, #T_a780d_row6_col17, #T_a780d_row7_col6, #T_a780d_row7_col17, #T_a780d_row8_col6, #T_a780d_row8_col17, #T_a780d_row9_col6, #T_a780d_row9_col17, #T_a780d_row10_col6, #T_a780d_row10_col14, #T_a780d_row10_col17, #T_a780d_row11_col6, #T_a780d_row11_col17 {
+  background-color: #440154;
+  color: #f1f1f1;
+}
+#T_a780d_row0_col4, #T_a780d_row0_col7, #T_a780d_row0_col8, #T_a780d_row0_col9, #T_a780d_row0_col12, #T_a780d_row1_col13, #T_a780d_row1_col14, #T_a780d_row2_col15, #T_a780d_row9_col13, #T_a780d_row10_col13, #T_a780d_row11_col2, #T_a780d_row11_col3, #T_a780d_row11_col5, #T_a780d_row11_col10, #T_a780d_row11_col11, #T_a780d_row11_col13, #T_a780d_row11_col16 {
+  background-color: #fde725;
+  color: #000000;
+}
+#T_a780d_row0_col5 {
+  background-color: #482576;
+  color: #f1f1f1;
+}
+#T_a780d_row0_col10 {
+  background-color: #443a83;
+  color: #f1f1f1;
+}
+#T_a780d_row0_col11 {
+  background-color: #414287;
+  color: #f1f1f1;
+}
+#T_a780d_row0_col13 {
+  background-color: #fbe723;
+  color: #000000;
+}
+#T_a780d_row0_col14 {
+  background-color: #2e6f8e;
+  color: #f1f1f1;
+}
+#T_a780d_row0_col15, #T_a780d_row6_col13 {
+  background-color: #f4e61e;
+  color: #000000;
+}
+#T_a780d_row1_col2, #T_a780d_row5_col16 {
+  background-color: #482173;
+  color: #f1f1f1;
+}
+#T_a780d_row1_col3 {
+  background-color: #48186a;
+  color: #f1f1f1;
+}
+#T_a780d_row1_col8 {
+  background-color: #1e9d89;
+  color: #f1f1f1;
+}
+#T_a780d_row1_col9 {
+  background-color: #d5e21a;
+  color: #000000;
+}
+#T_a780d_row1_col10, #T_a780d_row2_col16, #T_a780d_row5_col10, #T_a780d_row10_col4 {
+  background-color: #460a5d;
+  color: #f1f1f1;
+}
+#T_a780d_row1_col11 {
+  background-color: #470e61;
+  color: #f1f1f1;
+}
+#T_a780d_row1_col12 {
+  background-color: #218f8d;
+  color: #f1f1f1;
+}
+#T_a780d_row1_col15 {
+  background-color: #4cc26c;
+  color: #000000;
+}
+#T_a780d_row1_col16, #T_a780d_row3_col4, #T_a780d_row3_col7, #T_a780d_row4_col4, #T_a780d_row4_col7, #T_a780d_row5_col4, #T_a780d_row5_col7 {
+  background-color: #440256;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col2, #T_a780d_row2_col3 {
+  background-color: #433e85;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col5, #T_a780d_row4_col8 {
+  background-color: #482677;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col8 {
+  background-color: #2c738e;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col9 {
+  background-color: #44bf70;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col10 {
+  background-color: #277e8e;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col11 {
+  background-color: #25838e;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col12 {
+  background-color: #31668e;
+  color: #f1f1f1;
+}
+#T_a780d_row2_col13, #T_a780d_row4_col13 {
+  background-color: #90d743;
+  color: #000000;
+}
+#T_a780d_row2_col14 {
+  background-color: #e7e419;
+  color: #000000;
+}
+#T_a780d_row3_col2 {
+  background-color: #38588c;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col3, #T_a780d_row9_col12 {
+  background-color: #39568c;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col5 {
+  background-color: #2b758e;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col8 {
+  background-color: #2f6c8e;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col9, #T_a780d_row5_col9 {
+  background-color: #33628d;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col12 {
+  background-color: #3f4889;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col14 {
+  background-color: #d8e219;
+  color: #000000;
+}
+#T_a780d_row3_col15 {
+  background-color: #32658e;
+  color: #f1f1f1;
+}
+#T_a780d_row3_col16, #T_a780d_row11_col4 {
+  background-color: #471063;
+  color: #f1f1f1;
+}
+#T_a780d_row4_col2 {
+  background-color: #2d708e;
+  color: #f1f1f1;
+}
+#T_a780d_row4_col3 {
+  background-color: #2e6e8e;
+  color: #f1f1f1;
+}
+#T_a780d_row4_col5 {
+  background-color: #21a685;
+  color: #f1f1f1;
+}
+#T_a780d_row4_col12, #T_a780d_row5_col15, #T_a780d_row6_col11, #T_a780d_row7_col15 {
+  background-color: #471365;
+  color: #f1f1f1;
+}
+#T_a780d_row4_col14 {
+  background-color: #95d840;
+  color: #000000;
+}
+#T_a780d_row4_col15, #T_a780d_row9_col8 {
+  background-color: #472a7a;
+  color: #f1f1f1;
+}
+#T_a780d_row4_col16, #T_a780d_row6_col10, #T_a780d_row7_col8, #T_a780d_row8_col15 {
+  background-color: #481769;
+  color: #f1f1f1;
+}
+#T_a780d_row5_col2, #T_a780d_row11_col9 {
+  background-color: #25858e;
+  color: #f1f1f1;
+}
+#T_a780d_row5_col3 {
+  background-color: #297b8e;
+  color: #f1f1f1;
+}
+#T_a780d_row5_col5 {
+  background-color: #24878e;
+  color: #f1f1f1;
+}
+#T_a780d_row5_col8 {
+  background-color: #481f70;
+  color: #f1f1f1;
+}
+#T_a780d_row5_col11, #T_a780d_row6_col12, #T_a780d_row9_col7 {
+  background-color: #46075a;
+  color: #f1f1f1;
+}
+#T_a780d_row5_col13 {
+  background-color: #8ed645;
+  color: #000000;
+}
+#T_a780d_row5_col14 {
+  background-color: #7fd34e;
+  color: #000000;
+}
+#T_a780d_row6_col2 {
+  background-color: #1e9b8a;
+  color: #f1f1f1;
+}
+#T_a780d_row6_col3, #T_a780d_row7_col3 {
+  background-color: #1f968b;
+  color: #f1f1f1;
+}
+#T_a780d_row6_col4, #T_a780d_row6_col7, #T_a780d_row7_col4, #T_a780d_row7_col7, #T_a780d_row8_col7 {
+  background-color: #450457;
+  color: #f1f1f1;
+}
+#T_a780d_row6_col5 {
+  background-color: #46c06f;
+  color: #f1f1f1;
+}
+#T_a780d_row6_col9 {
+  background-color: #481a6c;
+  color: #f1f1f1;
+}
+#T_a780d_row6_col14 {
+  background-color: #56c667;
+  color: #000000;
+}
+#T_a780d_row6_col16 {
+  background-color: #472f7d;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col2 {
+  background-color: #2ab07f;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col5, #T_a780d_row11_col12 {
+  background-color: #218e8d;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col9, #T_a780d_row9_col10 {
+  background-color: #34618d;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col10 {
+  background-color: #482878;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col11 {
+  background-color: #482475;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col12 {
+  background-color: #482071;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col13 {
+  background-color: #f1e51d;
+  color: #000000;
+}
+#T_a780d_row7_col14, #T_a780d_row8_col14 {
+  background-color: #3aba76;
+  color: #f1f1f1;
+}
+#T_a780d_row7_col16, #T_a780d_row10_col8 {
+  background-color: #424086;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col2 {
+  background-color: #52c569;
+  color: #000000;
+}
+#T_a780d_row8_col3 {
+  background-color: #1fa188;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col4, #T_a780d_row8_col9, #T_a780d_row9_col4 {
+  background-color: #450559;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col5 {
+  background-color: #73d056;
+  color: #000000;
+}
+#T_a780d_row8_col8 {
+  background-color: #481c6e;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col10 {
+  background-color: #443b84;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col11 {
+  background-color: #453781;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col12, #T_a780d_row9_col15 {
+  background-color: #472c7a;
+  color: #f1f1f1;
+}
+#T_a780d_row8_col13 {
+  background-color: #eae51a;
+  color: #000000;
+}
+#T_a780d_row8_col16 {
+  background-color: #38598c;
+  color: #f1f1f1;
+}
+#T_a780d_row9_col2 {
+  background-color: #86d549;
+  color: #000000;
+}
+#T_a780d_row9_col3 {
+  background-color: #25ac82;
+  color: #f1f1f1;
+}
+#T_a780d_row9_col5 {
+  background-color: #70cf57;
+  color: #000000;
+}
+#T_a780d_row9_col9 {
+  background-color: #433d84;
+  color: #f1f1f1;
+}
+#T_a780d_row9_col11, #T_a780d_row10_col15 {
+  background-color: #365d8d;
+  color: #f1f1f1;
+}
+#T_a780d_row9_col14 {
+  background-color: #1fa287;
+  color: #f1f1f1;
+}
+#T_a780d_row9_col16 {
+  background-color: #297a8e;
+  color: #f1f1f1;
+}
+#T_a780d_row10_col2 {
+  background-color: #c2df23;
+  color: #000000;
+}
+#T_a780d_row10_col3 {
+  background-color: #65cb5e;
+  color: #000000;
+}
+#T_a780d_row10_col5 {
+  background-color: #93d741;
+  color: #000000;
+}
+#T_a780d_row10_col7 {
+  background-color: #46085c;
+  color: #f1f1f1;
+}
+#T_a780d_row10_col9 {
+  background-color: #3e4c8a;
+  color: #f1f1f1;
+}
+#T_a780d_row10_col10 {
+  background-color: #9dd93b;
+  color: #000000;
+}
+#T_a780d_row10_col11 {
+  background-color: #a8db34;
+  color: #000000;
+}
+#T_a780d_row10_col12 {
+  background-color: #54c568;
+  color: #000000;
+}
+#T_a780d_row10_col16 {
+  background-color: #3fbc73;
+  color: #f1f1f1;
+}
+#T_a780d_row11_col7 {
+  background-color: #471164;
+  color: #f1f1f1;
+}
+#T_a780d_row11_col8 {
+  background-color: #3b528b;
+  color: #f1f1f1;
+}
+#T_a780d_row11_col14 {
+  background-color: #2db27d;
+  color: #f1f1f1;
+}
+#T_a780d_row11_col15 {
+  background-color: #23898e;
+  color: #f1f1f1;
+}
+</style>
+<table id="T_a780d">
+  <thead>
+    <tr>
+      <th class="blank level0" >&nbsp;</th>
+      <th id="T_a780d_level0_col0" class="col_heading level0 col0" >version</th>
+      <th id="T_a780d_level0_col1" class="col_heading level0 col1" >d_sae</th>
+      <th id="T_a780d_level0_col2" class="col_heading level0 col2" >layer</th>
+      <th id="T_a780d_level0_col3" class="col_heading level0 col3" >kl_div_with_sae</th>
+      <th id="T_a780d_level0_col4" class="col_heading level0 col4" >kl_div_with_ablation</th>
+      <th id="T_a780d_level0_col5" class="col_heading level0 col5" >ce_loss_with_sae</th>
+      <th id="T_a780d_level0_col6" class="col_heading level0 col6" >ce_loss_without_sae</th>
+      <th id="T_a780d_level0_col7" class="col_heading level0 col7" >ce_loss_with_ablation</th>
+      <th id="T_a780d_level0_col8" class="col_heading level0 col8" >kl_div_score</th>
+      <th id="T_a780d_level0_col9" class="col_heading level0 col9" >ce_loss_score</th>
+      <th id="T_a780d_level0_col10" class="col_heading level0 col10" >l2_norm_in</th>
+      <th id="T_a780d_level0_col11" class="col_heading level0 col11" >l2_norm_out</th>
+      <th id="T_a780d_level0_col12" class="col_heading level0 col12" >l2_ratio</th>
+      <th id="T_a780d_level0_col13" class="col_heading level0 col13" >l0</th>
+      <th id="T_a780d_level0_col14" class="col_heading level0 col14" >l1</th>
+      <th id="T_a780d_level0_col15" class="col_heading level0 col15" >explained_variance</th>
+      <th id="T_a780d_level0_col16" class="col_heading level0 col16" >mse</th>
+      <th id="T_a780d_level0_col17" class="col_heading level0 col17" >total_tokens_evaluated</th>
+      <th id="T_a780d_level0_col18" class="col_heading level0 col18" >filepath</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <th id="T_a780d_level0_row0" class="row_heading level0 row0" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_0/metrics.json</th>
+      <td id="T_a780d_row0_col0" class="data row0 col0" >5</td>
+      <td id="T_a780d_row0_col1" class="data row0 col1" >128</td>
+      <td id="T_a780d_row0_col2" class="data row0 col2" >0</td>
+      <td id="T_a780d_row0_col3" class="data row0 col3" >0.002883</td>
+      <td id="T_a780d_row0_col4" class="data row0 col4" >3.094083</td>
+      <td id="T_a780d_row0_col5" class="data row0 col5" >3.601894</td>
+      <td id="T_a780d_row0_col6" class="data row0 col6" >3.599065</td>
+      <td id="T_a780d_row0_col7" class="data row0 col7" >6.694649</td>
+      <td id="T_a780d_row0_col8" class="data row0 col8" >0.999068</td>
+      <td id="T_a780d_row0_col9" class="data row0 col9" >0.999086</td>
+      <td id="T_a780d_row0_col10" class="data row0 col10" >29.933449</td>
+      <td id="T_a780d_row0_col11" class="data row0 col11" >29.719198</td>
+      <td id="T_a780d_row0_col12" class="data row0 col12" >0.993006</td>
+      <td id="T_a780d_row0_col13" class="data row0 col13" >31.999350</td>
+      <td id="T_a780d_row0_col14" class="data row0 col14" >61.198639</td>
+      <td id="T_a780d_row0_col15" class="data row0 col15" >0.977519</td>
+      <td id="T_a780d_row0_col16" class="data row0 col16" >13.276111</td>
+      <td id="T_a780d_row0_col17" class="data row0 col17" >6144.000000</td>
+      <td id="T_a780d_row0_col18" class="data row0 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_0/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row1" class="row_heading level0 row1" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_1/metrics.json</th>
+      <td id="T_a780d_row1_col0" class="data row1 col0" >5</td>
+      <td id="T_a780d_row1_col1" class="data row1 col1" >128</td>
+      <td id="T_a780d_row1_col2" class="data row1 col2" >1</td>
+      <td id="T_a780d_row1_col3" class="data row1 col3" >0.004242</td>
+      <td id="T_a780d_row1_col4" class="data row1 col4" >0.051053</td>
+      <td id="T_a780d_row1_col5" class="data row1 col5" >3.599821</td>
+      <td id="T_a780d_row1_col6" class="data row1 col6" >3.599065</td>
+      <td id="T_a780d_row1_col7" class="data row1 col7" >3.652537</td>
+      <td id="T_a780d_row1_col8" class="data row1 col8" >0.916919</td>
+      <td id="T_a780d_row1_col9" class="data row1 col9" >0.985857</td>
+      <td id="T_a780d_row1_col10" class="data row1 col10" >18.973736</td>
+      <td id="T_a780d_row1_col11" class="data row1 col11" >18.205364</td>
+      <td id="T_a780d_row1_col12" class="data row1 col12" >0.932765</td>
+      <td id="T_a780d_row1_col13" class="data row1 col13" >32.000000</td>
+      <td id="T_a780d_row1_col14" class="data row1 col14" >84.060181</td>
+      <td id="T_a780d_row1_col15" class="data row1 col15" >0.916917</td>
+      <td id="T_a780d_row1_col16" class="data row1 col16" >17.682375</td>
+      <td id="T_a780d_row1_col17" class="data row1 col17" >6144.000000</td>
+      <td id="T_a780d_row1_col18" class="data row1 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_1/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row2" class="row_heading level0 row2" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_2/metrics.json</th>
+      <td id="T_a780d_row2_col0" class="data row2 col0" >5</td>
+      <td id="T_a780d_row2_col1" class="data row2 col1" >128</td>
+      <td id="T_a780d_row2_col2" class="data row2 col2" >2</td>
+      <td id="T_a780d_row2_col3" class="data row2 col3" >0.006752</td>
+      <td id="T_a780d_row2_col4" class="data row2 col4" >0.058747</td>
+      <td id="T_a780d_row2_col5" class="data row2 col5" >3.602034</td>
+      <td id="T_a780d_row2_col6" class="data row2 col6" >3.599065</td>
+      <td id="T_a780d_row2_col7" class="data row2 col7" >3.645913</td>
+      <td id="T_a780d_row2_col8" class="data row2 col8" >0.885070</td>
+      <td id="T_a780d_row2_col9" class="data row2 col9" >0.936614</td>
+      <td id="T_a780d_row2_col10" class="data row2 col10" >49.106537</td>
+      <td id="T_a780d_row2_col11" class="data row2 col11" >47.976685</td>
+      <td id="T_a780d_row2_col12" class="data row2 col12" >0.912467</td>
+      <td id="T_a780d_row2_col13" class="data row2 col13" >31.984375</td>
+      <td id="T_a780d_row2_col14" class="data row2 col14" >82.676140</td>
+      <td id="T_a780d_row2_col15" class="data row2 col15" >0.981090</td>
+      <td id="T_a780d_row2_col16" class="data row2 col16" >28.213791</td>
+      <td id="T_a780d_row2_col17" class="data row2 col17" >6144.000000</td>
+      <td id="T_a780d_row2_col18" class="data row2 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_2/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row3" class="row_heading level0 row3" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_3/metrics.json</th>
+      <td id="T_a780d_row3_col0" class="data row3 col0" >5</td>
+      <td id="T_a780d_row3_col1" class="data row3 col1" >128</td>
+      <td id="T_a780d_row3_col2" class="data row3 col2" >3</td>
+      <td id="T_a780d_row3_col3" class="data row3 col3" >0.008532</td>
+      <td id="T_a780d_row3_col4" class="data row3 col4" >0.070592</td>
+      <td id="T_a780d_row3_col5" class="data row3 col5" >3.607723</td>
+      <td id="T_a780d_row3_col6" class="data row3 col6" >3.599065</td>
+      <td id="T_a780d_row3_col7" class="data row3 col7" >3.658678</td>
+      <td id="T_a780d_row3_col8" class="data row3 col8" >0.879137</td>
+      <td id="T_a780d_row3_col9" class="data row3 col9" >0.854756</td>
+      <td id="T_a780d_row3_col10" class="data row3 col10" >16.987318</td>
+      <td id="T_a780d_row3_col11" class="data row3 col11" >15.537837</td>
+      <td id="T_a780d_row3_col12" class="data row3 col12" >0.899518</td>
+      <td id="T_a780d_row3_col13" class="data row3 col13" >31.905111</td>
+      <td id="T_a780d_row3_col14" class="data row3 col14" >81.886444</td>
+      <td id="T_a780d_row3_col15" class="data row3 col15" >0.827494</td>
+      <td id="T_a780d_row3_col16" class="data row3 col16" >38.814980</td>
+      <td id="T_a780d_row3_col17" class="data row3 col17" >6144.000000</td>
+      <td id="T_a780d_row3_col18" class="data row3 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_3/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row4" class="row_heading level0 row4" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_4/metrics.json</th>
+      <td id="T_a780d_row4_col0" class="data row4 col0" >5</td>
+      <td id="T_a780d_row4_col1" class="data row4 col1" >128</td>
+      <td id="T_a780d_row4_col2" class="data row4 col2" >4</td>
+      <td id="T_a780d_row4_col3" class="data row4 col3" >0.010456</td>
+      <td id="T_a780d_row4_col4" class="data row4 col4" >0.063325</td>
+      <td id="T_a780d_row4_col5" class="data row4 col5" >3.611892</td>
+      <td id="T_a780d_row4_col6" class="data row4 col6" >3.599065</td>
+      <td id="T_a780d_row4_col7" class="data row4 col7" >3.660080</td>
+      <td id="T_a780d_row4_col8" class="data row4 col8" >0.834885</td>
+      <td id="T_a780d_row4_col9" class="data row4 col9" >0.789762</td>
+      <td id="T_a780d_row4_col10" class="data row4 col10" >17.251986</td>
+      <td id="T_a780d_row4_col11" class="data row4 col11" >15.434065</td>
+      <td id="T_a780d_row4_col12" class="data row4 col12" >0.879446</td>
+      <td id="T_a780d_row4_col13" class="data row4 col13" >31.984375</td>
+      <td id="T_a780d_row4_col14" class="data row4 col14" >78.371658</td>
+      <td id="T_a780d_row4_col15" class="data row4 col15" >0.780778</td>
+      <td id="T_a780d_row4_col16" class="data row4 col16" >50.818203</td>
+      <td id="T_a780d_row4_col17" class="data row4 col17" >6144.000000</td>
+      <td id="T_a780d_row4_col18" class="data row4 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_4/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row5" class="row_heading level0 row5" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_5/metrics.json</th>
+      <td id="T_a780d_row5_col0" class="data row5 col0" >5</td>
+      <td id="T_a780d_row5_col1" class="data row5 col1" >128</td>
+      <td id="T_a780d_row5_col2" class="data row5 col2" >5</td>
+      <td id="T_a780d_row5_col3" class="data row5 col3" >0.011618</td>
+      <td id="T_a780d_row5_col4" class="data row5 col4" >0.068505</td>
+      <td id="T_a780d_row5_col5" class="data row5 col5" >3.609259</td>
+      <td id="T_a780d_row5_col6" class="data row5 col6" >3.599065</td>
+      <td id="T_a780d_row5_col7" class="data row5 col7" >3.669386</td>
+      <td id="T_a780d_row5_col8" class="data row5 col8" >0.830411</td>
+      <td id="T_a780d_row5_col9" class="data row5 col9" >0.855026</td>
+      <td id="T_a780d_row5_col10" class="data row5 col10" >18.888968</td>
+      <td id="T_a780d_row5_col11" class="data row5 col11" >16.669851</td>
+      <td id="T_a780d_row5_col12" class="data row5 col12" >0.873767</td>
+      <td id="T_a780d_row5_col13" class="data row5 col13" >31.983725</td>
+      <td id="T_a780d_row5_col14" class="data row5 col14" >77.139160</td>
+      <td id="T_a780d_row5_col15" class="data row5 col15" >0.765372</td>
+      <td id="T_a780d_row5_col16" class="data row5 col16" >71.502914</td>
+      <td id="T_a780d_row5_col17" class="data row5 col17" >6144.000000</td>
+      <td id="T_a780d_row5_col18" class="data row5 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_5/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row6" class="row_heading level0 row6" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_6/metrics.json</th>
+      <td id="T_a780d_row6_col0" class="data row6 col0" >5</td>
+      <td id="T_a780d_row6_col1" class="data row6 col1" >128</td>
+      <td id="T_a780d_row6_col2" class="data row6 col2" >6</td>
+      <td id="T_a780d_row6_col3" class="data row6 col3" >0.014023</td>
+      <td id="T_a780d_row6_col4" class="data row6 col4" >0.075694</td>
+      <td id="T_a780d_row6_col5" class="data row6 col5" >3.614241</td>
+      <td id="T_a780d_row6_col6" class="data row6 col6" >3.599065</td>
+      <td id="T_a780d_row6_col7" class="data row6 col7" >3.676516</td>
+      <td id="T_a780d_row6_col8" class="data row6 col8" >0.814737</td>
+      <td id="T_a780d_row6_col9" class="data row6 col9" >0.804058</td>
+      <td id="T_a780d_row6_col10" class="data row6 col10" >21.466564</td>
+      <td id="T_a780d_row6_col11" class="data row6 col11" >18.890602</td>
+      <td id="T_a780d_row6_col12" class="data row6 col12" >0.875795</td>
+      <td id="T_a780d_row6_col13" class="data row6 col13" >31.998373</td>
+      <td id="T_a780d_row6_col14" class="data row6 col14" >74.617645</td>
+      <td id="T_a780d_row6_col15" class="data row6 col15" >0.754052</td>
+      <td id="T_a780d_row6_col16" class="data row6 col16" >97.221252</td>
+      <td id="T_a780d_row6_col17" class="data row6 col17" >6144.000000</td>
+      <td id="T_a780d_row6_col18" class="data row6 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_6/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row7" class="row_heading level0 row7" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_7/metrics.json</th>
+      <td id="T_a780d_row7_col0" class="data row7 col0" >5</td>
+      <td id="T_a780d_row7_col1" class="data row7 col1" >128</td>
+      <td id="T_a780d_row7_col2" class="data row7 col2" >7</td>
+      <td id="T_a780d_row7_col3" class="data row7 col3" >0.014005</td>
+      <td id="T_a780d_row7_col4" class="data row7 col4" >0.080486</td>
+      <td id="T_a780d_row7_col5" class="data row7 col5" >3.609832</td>
+      <td id="T_a780d_row7_col6" class="data row7 col6" >3.599065</td>
+      <td id="T_a780d_row7_col7" class="data row7 col7" >3.672712</td>
+      <td id="T_a780d_row7_col8" class="data row7 col8" >0.825992</td>
+      <td id="T_a780d_row7_col9" class="data row7 col9" >0.853803</td>
+      <td id="T_a780d_row7_col10" class="data row7 col10" >25.444439</td>
+      <td id="T_a780d_row7_col11" class="data row7 col11" >22.559195</td>
+      <td id="T_a780d_row7_col12" class="data row7 col12" >0.884380</td>
+      <td id="T_a780d_row7_col13" class="data row7 col13" >31.997885</td>
+      <td id="T_a780d_row7_col14" class="data row7 col14" >72.551300</td>
+      <td id="T_a780d_row7_col15" class="data row7 col15" >0.764806</td>
+      <td id="T_a780d_row7_col16" class="data row7 col16" >130.780823</td>
+      <td id="T_a780d_row7_col17" class="data row7 col17" >6144.000000</td>
+      <td id="T_a780d_row7_col18" class="data row7 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_7/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row8" class="row_heading level0 row8" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_8/metrics.json</th>
+      <td id="T_a780d_row8_col0" class="data row8 col0" >5</td>
+      <td id="T_a780d_row8_col1" class="data row8 col1" >128</td>
+      <td id="T_a780d_row8_col2" class="data row8 col2" >8</td>
+      <td id="T_a780d_row8_col3" class="data row8 col3" >0.014951</td>
+      <td id="T_a780d_row8_col4" class="data row8 col4" >0.087324</td>
+      <td id="T_a780d_row8_col5" class="data row8 col5" >3.615898</td>
+      <td id="T_a780d_row8_col6" class="data row8 col6" >3.599065</td>
+      <td id="T_a780d_row8_col7" class="data row8 col7" >3.680337</td>
+      <td id="T_a780d_row8_col8" class="data row8 col8" >0.828790</td>
+      <td id="T_a780d_row8_col9" class="data row8 col9" >0.792874</td>
+      <td id="T_a780d_row8_col10" class="data row8 col10" >30.250225</td>
+      <td id="T_a780d_row8_col11" class="data row8 col11" >26.934555</td>
+      <td id="T_a780d_row8_col12" class="data row8 col12" >0.888432</td>
+      <td id="T_a780d_row8_col13" class="data row8 col13" >31.996908</td>
+      <td id="T_a780d_row8_col14" class="data row8 col14" >72.559128</td>
+      <td id="T_a780d_row8_col15" class="data row8 col15" >0.768030</td>
+      <td id="T_a780d_row8_col16" class="data row8 col16" >183.634979</td>
+      <td id="T_a780d_row8_col17" class="data row8 col17" >6144.000000</td>
+      <td id="T_a780d_row8_col18" class="data row8 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_8/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row9" class="row_heading level0 row9" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_9/metrics.json</th>
+      <td id="T_a780d_row9_col0" class="data row9 col0" >5</td>
+      <td id="T_a780d_row9_col1" class="data row9 col1" >128</td>
+      <td id="T_a780d_row9_col2" class="data row9 col2" >9</td>
+      <td id="T_a780d_row9_col3" class="data row9 col3" >0.015919</td>
+      <td id="T_a780d_row9_col4" class="data row9 col4" >0.097589</td>
+      <td id="T_a780d_row9_col5" class="data row9 col5" >3.615844</td>
+      <td id="T_a780d_row9_col6" class="data row9 col6" >3.599065</td>
+      <td id="T_a780d_row9_col7" class="data row9 col7" >3.696245</td>
+      <td id="T_a780d_row9_col8" class="data row9 col8" >0.836876</td>
+      <td id="T_a780d_row9_col9" class="data row9 col9" >0.827333</td>
+      <td id="T_a780d_row9_col10" class="data row9 col10" >40.192413</td>
+      <td id="T_a780d_row9_col11" class="data row9 col11" >36.576889</td>
+      <td id="T_a780d_row9_col12" class="data row9 col12" >0.905792</td>
+      <td id="T_a780d_row9_col13" class="data row9 col13" >32.000000</td>
+      <td id="T_a780d_row9_col14" class="data row9 col14" >68.976295</td>
+      <td id="T_a780d_row9_col15" class="data row9 col15" >0.782096</td>
+      <td id="T_a780d_row9_col16" class="data row9 col16" >267.216095</td>
+      <td id="T_a780d_row9_col17" class="data row9 col17" >6144.000000</td>
+      <td id="T_a780d_row9_col18" class="data row9 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_9/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row10" class="row_heading level0 row10" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_10/metrics.json</th>
+      <td id="T_a780d_row10_col0" class="data row10 col0" >5</td>
+      <td id="T_a780d_row10_col1" class="data row10 col1" >128</td>
+      <td id="T_a780d_row10_col2" class="data row10 col2" >10</td>
+      <td id="T_a780d_row10_col3" class="data row10 col3" >0.019077</td>
+      <td id="T_a780d_row10_col4" class="data row10 col4" >0.126748</td>
+      <td id="T_a780d_row10_col5" class="data row10 col5" >3.616906</td>
+      <td id="T_a780d_row10_col6" class="data row10 col6" >3.599065</td>
+      <td id="T_a780d_row10_col7" class="data row10 col7" >3.708984</td>
+      <td id="T_a780d_row10_col8" class="data row10 col8" >0.849488</td>
+      <td id="T_a780d_row10_col9" class="data row10 col9" >0.837687</td>
+      <td id="T_a780d_row10_col10" class="data row10 col10" >81.756828</td>
+      <td id="T_a780d_row10_col11" class="data row10 col11" >78.834885</td>
+      <td id="T_a780d_row10_col12" class="data row10 col12" >0.960933</td>
+      <td id="T_a780d_row10_col13" class="data row10 col13" >32.000000</td>
+      <td id="T_a780d_row10_col14" class="data row10 col14" >48.323631</td>
+      <td id="T_a780d_row10_col15" class="data row10 col15" >0.819787</td>
+      <td id="T_a780d_row10_col16" class="data row10 col16" >443.108887</td>
+      <td id="T_a780d_row10_col17" class="data row10 col17" >6144.000000</td>
+      <td id="T_a780d_row10_col18" class="data row10 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_10/metrics.json</td>
+    </tr>
+    <tr>
+      <th id="T_a780d_level0_row11" class="row_heading level0 row11" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_11/metrics.json</th>
+      <td id="T_a780d_row11_col0" class="data row11 col0" >5</td>
+      <td id="T_a780d_row11_col1" class="data row11 col1" >128</td>
+      <td id="T_a780d_row11_col2" class="data row11 col2" >11</td>
+      <td id="T_a780d_row11_col3" class="data row11 col3" >0.024072</td>
+      <td id="T_a780d_row11_col4" class="data row11 col4" >0.173841</td>
+      <td id="T_a780d_row11_col5" class="data row11 col5" >3.620230</td>
+      <td id="T_a780d_row11_col6" class="data row11 col6" >3.599065</td>
+      <td id="T_a780d_row11_col7" class="data row11 col7" >3.783318</td>
+      <td id="T_a780d_row11_col8" class="data row11 col8" >0.861531</td>
+      <td id="T_a780d_row11_col9" class="data row11 col9" >0.885130</td>
+      <td id="T_a780d_row11_col10" class="data row11 col10" >92.906296</td>
+      <td id="T_a780d_row11_col11" class="data row11 col11" >88.290031</td>
+      <td id="T_a780d_row11_col12" class="data row11 col12" >0.932119</td>
+      <td id="T_a780d_row11_col13" class="data row11 col13" >32.000000</td>
+      <td id="T_a780d_row11_col14" class="data row11 col14" >71.255219</td>
+      <td id="T_a780d_row11_col15" class="data row11 col15" >0.861334</td>
+      <td id="T_a780d_row11_col16" class="data row11 col16" >635.459473</td>
+      <td id="T_a780d_row11_col17" class="data row11 col17" >6144.000000</td>
+      <td id="T_a780d_row11_col18" class="data row11 col18" >OAI_GPT2Small_v5_128k_resid_delta_mlp/v5_128k_layer_11/metrics.json</td>
+    </tr>
+  </tbody>
+</table>

v5_128k_layer_0/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.0.hook_mlp_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"}

v5_128k_layer_0/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.0028826396446675062, "metrics/kl_div_with_ablation": 3.094083309173584, "metrics/ce_loss_with_sae": 3.6018943786621094, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 6.694648742675781, "metrics/kl_div_score": 0.999068338064421, "metrics/ce_loss_score": 0.9990858623204987, "metrics/l2_norm_in": 29.933448791503906, "metrics/l2_norm_out": 29.71919822692871, "metrics/l2_ratio": 0.9930059909820557, "metrics/l0": 31.99934959411621, "metrics/l1": 61.198638916015625, "metrics/explained_variance": 0.9775193929672241, "metrics/mse": 13.276110649108887, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_0/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e313db224266d10978eb24b4c3e5190d0453204822c65075e077702d52a586c
+size 805834048

v5_128k_layer_0/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f5401432bfa99a5fff073491602e63988a3956cad6fad7b6bec93e25a44cddf
+size 524368

v5_128k_layer_1/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.1.hook_mlp_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"}

v5_128k_layer_1/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.004241525661200285, "metrics/kl_div_with_ablation": 0.05105271190404892, "metrics/ce_loss_with_sae": 3.599820852279663, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6525371074676514, "metrics/kl_div_score": 0.9169187002411934, "metrics/ce_loss_score": 0.9858569645086499, "metrics/l2_norm_in": 18.973735809326172, "metrics/l2_norm_out": 18.205364227294922, "metrics/l2_ratio": 0.9327648878097534, "metrics/l0": 32.0, "metrics/l1": 84.0601806640625, "metrics/explained_variance": 0.9169172048568726, "metrics/mse": 17.682374954223633, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_1/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46095a456b1e23d7fdd8688853bfcc6797bdc67c5aecc76d61662b3512201582
+size 805834048

v5_128k_layer_1/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70d5f5cfc52d74f2bf6737a5d7fd175f425314dcdd34be553e0ed5058812e1ff
+size 524368

v5_128k_layer_10/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.10.hook_mlp_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"}

v5_128k_layer_10/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.019077030941843987, "metrics/kl_div_with_ablation": 0.12674781680107117, "metrics/ce_loss_with_sae": 3.616905927658081, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.708984136581421, "metrics/kl_div_score": 0.8494882876619082, "metrics/ce_loss_score": 0.8376872955691096, "metrics/l2_norm_in": 81.75682830810547, "metrics/l2_norm_out": 78.83488464355469, "metrics/l2_ratio": 0.9609333872795105, "metrics/l0": 32.0, "metrics/l1": 48.323631286621094, "metrics/explained_variance": 0.819786787033081, "metrics/mse": 443.10888671875, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_10/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4290e02f0d43cf57ce0050987a6ae00af766f43a47c013c77c63a8745b89b43
+size 805834048

v5_128k_layer_10/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e42a413071da022c6c2e5ba5fe561d358348b76a0126d5f313187b5243e1e8b1
+size 524368

v5_128k_layer_11/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.11.hook_mlp_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"}

v5_128k_layer_11/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.024071572348475456, "metrics/kl_div_with_ablation": 0.17384076118469238, "metrics/ce_loss_with_sae": 3.620229721069336, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.783318281173706, "metrics/kl_div_score": 0.8615309080308198, "metrics/ce_loss_score": 0.8851304838409143, "metrics/l2_norm_in": 92.90629577636719, "metrics/l2_norm_out": 88.29003143310547, "metrics/l2_ratio": 0.9321191310882568, "metrics/l0": 32.0, "metrics/l1": 71.25521850585938, "metrics/explained_variance": 0.8613338470458984, "metrics/mse": 635.45947265625, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_11/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77f326d4427af047f719a0e8328554ca67bfceaddc28627dc7c97f691509ee28
+size 805834048

v5_128k_layer_11/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25f73118e68ddef034d3e396b3689cb81d3a2d8a1352cfa76175d62b64b31c1d
+size 524368

v5_128k_layer_2/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.2.hook_mlp_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"}

v5_128k_layer_2/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.006751821842044592, "metrics/kl_div_with_ablation": 0.05874736234545708, "metrics/ce_loss_with_sae": 3.602034091949463, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6459126472473145, "metrics/kl_div_score": 0.8850702129852012, "metrics/ce_loss_score": 0.9366141632102598, "metrics/l2_norm_in": 49.106536865234375, "metrics/l2_norm_out": 47.9766845703125, "metrics/l2_ratio": 0.912467360496521, "metrics/l0": 31.984375, "metrics/l1": 82.67613983154297, "metrics/explained_variance": 0.981090247631073, "metrics/mse": 28.213790893554688, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_2/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d54e5970d48359a47600a92f1f073c0b5517d85adb3da218ac6919d76477c5a
+size 805834048

v5_128k_layer_2/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb0cb9d36290181dd877afed97aa7149ae33b6157465a83f7d8a53ffc0fb1cb9
+size 524368

v5_128k_layer_3/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.3.hook_mlp_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"}

v5_128k_layer_3/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.008531898260116577, "metrics/kl_div_with_ablation": 0.07059153914451599, "metrics/ce_loss_with_sae": 3.6077229976654053, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.658677577972412, "metrics/kl_div_score": 0.8791370982484182, "metrics/ce_loss_score": 0.8547563341132242, "metrics/l2_norm_in": 16.98731803894043, "metrics/l2_norm_out": 15.537837028503418, "metrics/l2_ratio": 0.8995180130004883, "metrics/l0": 31.90511131286621, "metrics/l1": 81.88644409179688, "metrics/explained_variance": 0.8274940252304077, "metrics/mse": 38.814979553222656, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_3/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39589625da53203638c9c34d238ef6c6548db867b6fd9c99d3a46bea77403159
+size 805834048

v5_128k_layer_3/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:539e71813f11254b3c12d743073e0c34052aa3d41a3f5067b6c28e216b0f8024
+size 524368

v5_128k_layer_4/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.4.hook_mlp_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"}

v5_128k_layer_4/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.010455957613885403, "metrics/kl_div_with_ablation": 0.06332532316446304, "metrics/ce_loss_with_sae": 3.6118924617767334, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6600804328918457, "metrics/kl_div_score": 0.8348850492759414, "metrics/ce_loss_score": 0.7897616042575971, "metrics/l2_norm_in": 17.251985549926758, "metrics/l2_norm_out": 15.434064865112305, "metrics/l2_ratio": 0.8794459700584412, "metrics/l0": 31.984375, "metrics/l1": 78.37165832519531, "metrics/explained_variance": 0.780777633190155, "metrics/mse": 50.81820297241211, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_4/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6487cb7e7a65f9bcfcea94c2214ab1408192e1d1b24690aa5fb2837218e809e
+size 805834048

v5_128k_layer_4/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7ed6462fe0e62b4cacfb2b21d3e3e18668eba46a042911a04f316c7ece41086
+size 524368

v5_128k_layer_5/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.5.hook_mlp_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"}

v5_128k_layer_5/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.011617729440331459, "metrics/kl_div_with_ablation": 0.06850520521402359, "metrics/ce_loss_with_sae": 3.6092593669891357, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6693859100341797, "metrics/kl_div_score": 0.8304109971784566, "metrics/ce_loss_score": 0.8550257841186103, "metrics/l2_norm_in": 18.888967514038086, "metrics/l2_norm_out": 16.669851303100586, "metrics/l2_ratio": 0.8737667798995972, "metrics/l0": 31.98372459411621, "metrics/l1": 77.13916015625, "metrics/explained_variance": 0.7653716802597046, "metrics/mse": 71.50291442871094, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_5/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14b24eb29046b3f305fadb6868e2afaf4f42d78ab4666a5240470691141cd301
+size 805834048

v5_128k_layer_5/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59744fae94a6eba1e2e064ffda767c2836208cc63554ab1c2e567eff78593263
+size 524368

v5_128k_layer_6/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.6.hook_mlp_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"}

v5_128k_layer_6/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.014023388735949993, "metrics/kl_div_with_ablation": 0.07569437474012375, "metrics/ce_loss_with_sae": 3.6142406463623047, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.67651629447937, "metrics/kl_div_score": 0.814736712152052, "metrics/ce_loss_score": 0.8040577979166154, "metrics/l2_norm_in": 21.466564178466797, "metrics/l2_norm_out": 18.890602111816406, "metrics/l2_ratio": 0.8757946491241455, "metrics/l0": 31.99837303161621, "metrics/l1": 74.61764526367188, "metrics/explained_variance": 0.754051685333252, "metrics/mse": 97.22125244140625, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_6/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93f5c45db73104697a3c8e8d708639978852ce69fe715da1c073d5de8124baca
+size 805834048

v5_128k_layer_6/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a872de1296a7b9cd559fdc57ebc575f256bd29f2d08fcaea2abeab10751bb58c
+size 524368

v5_128k_layer_7/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.7.hook_mlp_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"}

v5_128k_layer_7/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.014005273580551147, "metrics/kl_div_with_ablation": 0.08048636466264725, "metrics/ce_loss_with_sae": 3.6098315715789795, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6727118492126465, "metrics/kl_div_score": 0.8259919721899077, "metrics/ce_loss_score": 0.8538033467249813, "metrics/l2_norm_in": 25.444438934326172, "metrics/l2_norm_out": 22.559194564819336, "metrics/l2_ratio": 0.8843803405761719, "metrics/l0": 31.99788475036621, "metrics/l1": 72.55130004882812, "metrics/explained_variance": 0.7648060321807861, "metrics/mse": 130.78082275390625, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_7/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:caeb6361e42d377cadc85c5fb6d1b488bd9dc4db0223133903ee4df02a78c73f
+size 805834048

v5_128k_layer_7/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5efc7e378076c3fc51ac2e5a98e235b2b3a3aae78305d9055847074ee7c9d699
+size 524368

v5_128k_layer_8/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.8.hook_mlp_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"}

v5_128k_layer_8/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.0149507075548172, "metrics/kl_div_with_ablation": 0.08732372522354126, "metrics/ce_loss_with_sae": 3.6158981323242188, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6803367137908936, "metrics/kl_div_score": 0.828789856175459, "metrics/ce_loss_score": 0.7928743252757569, "metrics/l2_norm_in": 30.250225067138672, "metrics/l2_norm_out": 26.934555053710938, "metrics/l2_ratio": 0.8884322047233582, "metrics/l0": 31.99690818786621, "metrics/l1": 72.55912780761719, "metrics/explained_variance": 0.7680301070213318, "metrics/mse": 183.63497924804688, "metrics/total_tokens_evaluated": 6144}

v5_128k_layer_8/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bae64cc48c181215c8dd40ca14bb1163dc9ae73ab18296d7b49e7ffe84729485
+size 805834048

v5_128k_layer_8/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe8f8bee61f208f771efbcdd8188b56e7a0eb74a32775d68c1adee78a27c3a8e
+size 524368

v5_128k_layer_9/cfg.json

@@ -0,0 +1 @@
+{"architecture": "standard", "d_in": 768, "d_sae": 131072, "dtype": "torch.float32", "device": "cuda", "model_name": "gpt2-small", "hook_name": "blocks.9.hook_mlp_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"}

v5_128k_layer_9/metrics.json

@@ -0,0 +1 @@
+{"metrics/kl_div_with_sae": 0.01591905765235424, "metrics/kl_div_with_ablation": 0.09758877754211426, "metrics/ce_loss_with_sae": 3.615844488143921, "metrics/ce_loss_without_sae": 3.599064588546753, "metrics/ce_loss_with_ablation": 3.6962451934814453, "metrics/kl_div_score": 0.8368761444369522, "metrics/ce_loss_score": 0.827332834484366, "metrics/l2_norm_in": 40.192413330078125, "metrics/l2_norm_out": 36.57688903808594, "metrics/l2_ratio": 0.9057919383049011, "metrics/l0": 32.0, "metrics/l1": 68.9762954711914, "metrics/explained_variance": 0.7820959091186523, "metrics/mse": 267.2160949707031, "metrics/total_tokens_evaluated": 6144}