Upload benchmark_24h.json with huggingface_hub

benchmark_24h.json

@@ -1,280 +1,344 @@
 {
-  "step": 122743,
+  "step": 105500,
   "crystallization": {
-    "slot0_ce": 1.437141352891922,
-    "slot0_ppl": 4.208647566158646,
-    "slot0_acc": 0.6439453363418579,
-    "causal_lm_ce": 1.2504938185214995,
-    "causal_lm_ppl": 3.492066979102496,
-    "causal_lm_acc": 0.6677703201770783,
-    "mean_plan_pause": 0.05949237179011106,
+    "slot0_ce": 1.6622225046157837,
+    "slot0_ppl": 5.27101268595611,
+    "slot0_acc": 0.595507800579071,
+    "causal_lm_ce": 1.5311720550060273,
+    "causal_lm_ppl": 4.623592752827325,
+    "causal_lm_acc": 0.611477792263031,
+    "mean_plan_pause": 0.5485862284898758,
     "per_slot_acc": [
-      0.6439453363418579,
-      0.47968751192092896,
-      0.3623046875,
-      0.2783203125,
-      0.21230469644069672,
-      0.16660156846046448,
+      0.595507800579071,
+      0.4371093809604645,
+      0.3169921934604645,
+      0.23476561903953552,
+      0.18203124403953552,
       0.13750000298023224,
-      0.11640625447034836,
-      0.10468750447034836,
-      0.09746094048023224,
-      0.09121093899011612,
-      0.08476562798023224,
-      0.08730468899011612,
-      0.08632812649011612,
-      0.08242187649011612,
+      0.05156249925494194,
+      0.05546874925494194,
+      0.06875000149011612,
+      0.05917968600988388,
+      0.05234374850988388,
+      0.05996093899011612,
+      0.06523437798023224,
+      0.0546875,
+      0.0458984375,
       0.07988281548023224,
-      0.080078125,
-      0.07539062947034836,
-      0.07421875,
-      0.07148437947034836,
-      0.06894531100988388,
-      0.07500000298023224,
-      0.07011719048023224,
-      0.07050781697034836,
-      0.06992187350988388,
-      0.07265625149011612,
-      0.0712890625,
-      0.07070312649011612,
+      0.05820312350988388,
+      0.0625,
+      0.06621094048023224,
+      0.06718750298023224,
+      0.06171875074505806,
+      0.068359375,
+      0.06113281100988388,
+      0.06035156175494194,
+      0.06640625,
       0.07089843600988388,
-      0.06972656399011612,
-      0.06972656399011612,
-      0.07050781697034836,
-      0.06953125447034836,
-      0.0703125,
-      0.07109375298023224,
-      0.06972656399011612,
-      0.06875000149011612,
-      0.06914062798023224,
-      0.0732683315873146,
-      0.07192770391702652,
-      0.07155468314886093,
-      0.07182707637548447,
-      0.07598132640123367,
-      0.07604477554559708,
-      0.07479842752218246,
-      0.07520607113838196,
-      0.07187029719352722,
-      0.076015405356884,
-      0.07749485969543457,
-      0.07287924736738205,
-      0.07080724090337753,
-      0.0724722370505333,
-      0.07274691760540009,
-      0.07342890650033951,
-      0.07258143275976181,
-      0.07091822475194931,
-      0.06911274045705795,
-      0.0738527923822403,
-      0.07288306951522827,
-      0.07187815755605698,
-      0.07096120715141296,
-      0.06949564069509506,
-      0.07068412005901337,
-      0.07690208405256271
+      0.06093750149011612,
+      0.06230468675494194,
+      0.06798943132162094,
+      0.06371356546878815,
+      0.068232960999012,
+      0.0657910630106926,
+      0.065426304936409,
+      0.06768668442964554,
+      0.0659736767411232,
+      0.07201062887907028,
+      0.07096768170595169,
+      0.0682436004281044,
+      0.07242675870656967,
+      0.07701095193624496,
+      0.076072558760643,
+      0.07489804923534393,
+      0.07112248986959457,
+      0.06749451160430908,
+      0.07228265702724457,
+      0.0816536694765091,
+      0.06897489726543427,
+      0.07450846582651138,
+      0.0674772560596466,
+      0.06346741318702698,
+      0.06750796735286713,
+      0.06961502134799957,
+      0.06840070337057114,
+      0.05763889476656914,
+      0.06445814669132233,
+      0.07262647897005081,
+      0.05406894534826279,
+      0.06724672764539719,
+      0.070048026740551,
+      0.06905129551887512,
+      0.06587334722280502,
+      0.06987018138170242,
+      0.05123468488454819,
+      0.0578719861805439,
+      0.059157419949769974,
+      0.049347348511219025,
+      0.053371310234069824,
+      0.047219909727573395,
+      0.05238068848848343,
+      0.06489019095897675,
+      0.0609363317489624,
+      0.05877644568681717,
+      0.07479767501354218,
+      0.0562639944255352,
+      0.04465726017951965,
+      0.044268302619457245,
+      0.047840412706136703,
+      0.05366982892155647,
+      0.04312526807188988,
+      0.05177409201860428,
+      0.04654914140701294,
+      0.05816614627838135,
+      0.052413154393434525,
+      0.045634377747774124,
+      0.04146602377295494,
+      0.04622279852628708,
+      0.05330605432391167,
+      0.05263037607073784,
+      0.04560557380318642,
+      0.03866346925497055,
+      0.05465855449438095,
+      0.043680835515260696,
+      0.03889264911413193,
+      0.025178994983434677,
+      0.03811141476035118,
+      0.03958333283662796
     ],
     "per_slot_entropy": [
-      0.4764425456523895,
-      0.7842304706573486,
-      1.137978434562683,
-      1.49298095703125,
-      1.8444280624389648,
-      2.1730051040649414,
-      2.450309991836548,
-      2.708646535873413,
-      2.9538815021514893,
-      3.1541783809661865,
-      3.3763020038604736,
-      3.5716769695281982,
-      3.7822368144989014,
-      4.006913661956787,
-      4.228540897369385,
-      4.45133113861084,
-      4.690694332122803,
-      4.9508562088012695,
-      5.177692890167236,
-      5.426706314086914,
-      5.657607555389404,
-      5.869189262390137,
-      6.065157413482666,
-      6.2689208984375,
-      6.427663326263428,
-      6.592019557952881,
-      6.739131450653076,
-      6.870829105377197,
-      6.988106727600098,
-      7.094132900238037,
-      7.1884074211120605,
-      7.283900737762451,
-      7.358842372894287,
-      7.436470031738281,
-      7.50555944442749,
-      7.568306922912598,
-      7.623337745666504,
-      7.672909736633301,
-      7.717443943023682,
-      7.758584499359131,
-      7.795356273651123,
-      7.8370866775512695,
-      7.8670549392700195,
-      7.894341945648193,
-      7.925351619720459,
-      7.947883605957031,
-      7.97347354888916,
-      7.992164134979248,
-      8.013349533081055,
-      8.032417297363281,
-      8.046835899353027,
-      8.063529014587402,
-      8.078781127929688,
-      8.092826843261719,
-      8.105636596679688,
-      8.117079734802246,
-      8.127867698669434,
-      8.138626098632812,
-      8.147955894470215,
-      8.156766891479492,
-      8.16491985321045,
-      8.171956062316895,
-      8.17918586730957,
-      8.18935775756836
+      0.4674500524997711,
+      0.7802783250808716,
+      1.0742617845535278,
+      1.3449093103408813,
+      1.5819367170333862,
+      2.184095621109009,
+      3.922560214996338,
+      4.224471092224121,
+      4.306208610534668,
+      4.356972694396973,
+      4.438023567199707,
+      4.515223503112793,
+      4.571393966674805,
+      4.668587684631348,
+      4.735049724578857,
+      4.2593464851379395,
+      4.853049278259277,
+      4.91046142578125,
+      4.888455867767334,
+      5.010046005249023,
+      5.053281784057617,
+      5.085631370544434,
+      5.137650489807129,
+      5.184286594390869,
+      5.218418598175049,
+      5.265626907348633,
+      5.3011674880981445,
+      5.348525047302246,
+      5.383389949798584,
+      5.4296746253967285,
+      5.4691548347473145,
+      5.507866859436035,
+      5.54647159576416,
+      5.58273458480835,
+      5.6498918533325195,
+      5.65679407119751,
+      5.464327335357666,
+      5.03488826751709,
+      5.0812788009643555,
+      5.130118370056152,
+      5.1806721687316895,
+      5.2379255294799805,
+      5.291325092315674,
+      5.337025165557861,
+      5.3868303298950195,
+      5.43253755569458,
+      5.477930545806885,
+      5.523505210876465,
+      5.577084541320801,
+      5.620858192443848,
+      5.66721773147583,
+      5.712264060974121,
+      5.75246000289917,
+      5.796233177185059,
+      5.838044166564941,
+      5.881399154663086,
+      5.922351837158203,
+      5.967439651489258,
+      6.005418300628662,
+      6.043013095855713,
+      6.089051246643066,
+      6.130578994750977,
+      6.172563552856445,
+      6.218247413635254,
+      6.254241943359375,
+      6.296483993530273,
+      6.336541175842285,
+      6.369558811187744,
+      6.403321743011475,
+      6.444928169250488,
+      6.493714809417725,
+      6.5418291091918945,
+      6.589001655578613,
+      6.620169639587402,
+      6.671649932861328,
+      6.692941188812256,
+      6.7116851806640625,
+      6.733679294586182,
+      6.764509677886963,
+      6.778643608093262,
+      6.8155837059021,
+      6.829553127288818,
+      6.846144199371338,
+      6.893029689788818,
+      6.917045593261719,
+      6.964493751525879,
+      7.074467658996582,
+      7.11127233505249,
+      7.145449161529541,
+      7.160573482513428,
+      7.165871620178223,
+      7.1802978515625,
+      7.197512626647949,
+      7.201399803161621,
+      7.223245143890381,
+      7.237560272216797
     ]
   },
   "commonsense": {
-    "commonsense_lm_acc": 1.0,
-    "commonsense_slot0_acc": 0.8125,
-    "mean_logprob_margin": 2.5752472206950188,
+    "commonsense_lm_acc": 0.9375,
+    "commonsense_slot0_acc": 0.875,
+    "mean_logprob_margin": 2.349940698593855,
     "num_pairs": 16,
     "details": [
       {
         "category": "emotion_cause",
         "correct": true,
-        "logprob_good": -1.046,
-        "logprob_bad": -3.206
+        "logprob_good": -1.229,
+        "logprob_bad": -3.342
       },
       {
         "category": "emotion_cause",
         "correct": true,
-        "logprob_good": -0.773,
-        "logprob_bad": -5.345
+        "logprob_good": -0.904,
+        "logprob_bad": -5.241
       },
       {
         "category": "physical_weight",
         "correct": true,
-        "logprob_good": -1.497,
-        "logprob_bad": -3.835
+        "logprob_good": -1.881,
+        "logprob_bad": -3.79
       },
       {
         "category": "physical_temperature",
         "correct": true,
-        "logprob_good": -0.876,
-        "logprob_bad": -3.949
+        "logprob_good": -0.897,
+        "logprob_bad": -2.812
       },
       {
         "category": "physical_temperature",
         "correct": true,
-        "logprob_good": -2.276,
-        "logprob_bad": -5.165
+        "logprob_good": -2.745,
+        "logprob_bad": -4.798
       },
       {
         "category": "physical_fragile",
         "correct": true,
-        "logprob_good": -1.173,
-        "logprob_bad": -4.762
+        "logprob_good": -2.413,
+        "logprob_bad": -4.861
       },
       {
         "category": "object_use",
         "correct": true,
-        "logprob_good": -3.329,
-        "logprob_bad": -3.461
+        "logprob_good": -3.11,
+        "logprob_bad": -4.776
       },
       {
         "category": "object_use",
         "correct": true,
-        "logprob_good": -1.659,
-        "logprob_bad": -3.931
+        "logprob_good": -1.955,
+        "logprob_bad": -4.411
       },
       {
         "category": "biological_need",
         "correct": true,
-        "logprob_good": -1.334,
-        "logprob_bad": -3.579
+        "logprob_good": -1.401,
+        "logprob_bad": -3.418
       },
       {
         "category": "biological_need",
-        "correct": true,
-        "logprob_good": -2.743,
-        "logprob_bad": -2.972
+        "correct": false,
+        "logprob_good": -3.454,
+        "logprob_bad": -3.383
       },
       {
         "category": "entity_tracking",
         "correct": true,
-        "logprob_good": -0.955,
-        "logprob_bad": -5.76
+        "logprob_good": -0.597,
+        "logprob_bad": -5.734
       },
       {
         "category": "entity_tracking",
         "correct": true,
-        "logprob_good": -2.175,
-        "logprob_bad": -5.094
+        "logprob_good": -2.245,
+        "logprob_bad": -4.326
       },
       {
         "category": "danger_reaction",
         "correct": true,
-        "logprob_good": -0.906,
-        "logprob_bad": -5.322
+        "logprob_good": -1.211,
+        "logprob_bad": -5.542
       },
       {
         "category": "social_repair",
         "correct": true,
-        "logprob_good": -1.188,
-        "logprob_bad": -4.079
+        "logprob_good": -1.214,
+        "logprob_bad": -3.966
       },
       {
         "category": "spatial_logic",
         "correct": true,
-        "logprob_good": -1.458,
-        "logprob_bad": -3.383
+        "logprob_good": -1.894,
+        "logprob_bad": -3.414
       },
       {
         "category": "weather_visibility",
         "correct": true,
-        "logprob_good": -1.864,
-        "logprob_bad": -2.612
+        "logprob_good": -2.234,
+        "logprob_bad": -3.167
       }
     ]
   },
   "generation_quality": {
-    "prefix_word_order_kl": 3.6741719245910645,
-    "distinct_2": 0.7803174059310821,
-    "distinct_3": 0.8672925871374969,
-    "repeat_4gram_rate": 0.09066065924220516,
+    "prefix_word_order_kl": 4.045264522234599,
+    "distinct_2": 0.8936134700745502,
+    "distinct_3": 0.9851329961647938,
+    "repeat_4gram_rate": 0.0024509803921568627,
     "prompt_entity_retention": 1.0,
     "samples": [
       {
         "prompt": "Once upon a time, Lily and a little bunny walked into the green forest to find the golden key near the locked treasure chest. Suddenly a big lion",
-        "continuation": "appeared and roared, \"I'm the king of the forest!\" Lily was scared and said, \"No, you can't! I'm the king of the forest!\" The lion roared, \"I'm the king of the forest! I'm the king of the forest!\" Lily was very scared and said, \"Please, I, I'm the king of the forest!\" The lion said, \""
+        "continuation": "appeared and said, \"Hello, little bunny! I have a special key to help you find the treasure. The key is very important to help, and it's to be kind to others. I will help you, but you must be very kind and help others too.\" The little bunny was very excited, so he listened to the little bunny's help. He followed the key to the treasure and"
       },
       {
         "prompt": "Once upon a time, there was a little robot named Beep who lived in a noisy workshop. One rainy morning, Beep found a tiny lost kitten by the door. Beep",
-        "continuation": "said, \"Oh no, little kitten! I need to help you find your home.\" The kitten was scared and said, \"Please,, I am lost. Can you help me?\" Beep thought for a moment and said, \"Yes, I will help you. Let's go find your home together.\" They walked and walked until they found a's house. The kitten was so happy and"
+        "continuation": "wanted to help the kitten, so he decided to help it. Beep said, \"Don't worry, little kitten. I will help you find your way home and help you.\" BeBe be very kind and let the kitten follow him. The little kitten was a little scared, so they walked and walked until they found the little kitten. The little puppy was very happy and said, \"Thank"
       },
       {
         "prompt": "Sam and his sister Mia built a huge sandcastle on the beach near the waves. Suddenly, a big wave splashed toward their castle. Sam quickly",
-        "continuation": "ran to the sand and and shouted, \"Look, Mia! A huge wave!\" Mia was and said, \"Let's go it!\" Sam ran to the sand and and they both ran to the sand.. They laughed and shouted as they made the sandcastle bigger and bigger. Suddenly, a huge wave came and and the sand was gone. Sam and Mia were sad. They looked around and"
+        "continuation": "ran to Mia and said, \ufffdLet\ufffdThat\ufffd go!\ufffd. Mia was so excited to see the wave that she ran towards the waves. She jumped into the sand and splashed in it. It was so hard that it was too hard and too hard. Sam\ufffd was getting hurt, so he ran to Mia\ufffd He was in pain. Mia hugged him and said, \ufffdDon\ufffd"
       },
       {
         "prompt": "In a quiet village, an old baker dropped his round loaf of bread and it rolled out the door down the hill. A hungry puppy saw the bread and",
-        "continuation": "ran to it. He barked and wagged his tail, but the puppy was too fast. He stamped his paws and ran away. The puppy chased and ran until he was far away. He was so tired that he couldn't see the.. He was to to the the puppy and he was to to the.. But the puppy was too fast and"
+        "continuation": "ran towards it. He was so excited to have the bread, but he was also very hungry. The old one day, the puppy decided to eat the bread with a little bit of his little. He ate it and it was so sweet that he could not stop eating it. The old, old one day, the little puppy had to go home. He was so hungry that he forgot about eating"
       },
       {
         "prompt": "Lucy had a box of bright crayons, but she could not find the blue crayon to color the sky. She looked under her bed and",
-        "continuation": "in her toy box, but it was not there. She asked her mom, \"Mom, where is my blue crayon?\" Her mom said, \"I don't know, Lucy. Maybe it's in the living room.\" So went to the living room and looked around. She saw the crayon on the table, but it was not there. She looked to the living room,"
+        "continuation": "in her toy box, but it was not there. She looked under her bed, in the toy box, and in the living room. She could not find her crayon anywhere. She started to cry. Mommy heard her cry and came to see. She saw Lucy crying and asked, \"What's wrong, Lucy?\" Lucy showed her the red crayon. Mommy said,"
       },
       {
         "prompt": "One winter day, Max made a snowman in the yard with a carrot nose. When the warm sun came out in the afternoon, the snowman began to",
-        "continuation": "melt. Max was so sad. He ran to his mom and said, \"Mommy, my snowman is gone!\" His mom smiled and said, \"Don't worry, Max. We can make a new snowman for you.\" Max smiled and and his mom got some supplies. They put them  the snowman in the snowman and added a"
+        "continuation": "melt. Max was so sad that he had to find a way to make the snowman melt. Max looked around and saw a big tree with a hole in it. He had an idea! He ran to the tree and got a big stick. He threw the stick into the hole and it made a big snowman! The snowman was so happy that he had made the snowman melt."
       }
     ]
   },
@@ -282,29 +346,29 @@
     {
       "bake_steps": 16,
       "prompt_overlap": 0,
-      "first_token_acc": 0.1875,
-      "token_acc": 0.032552083333333336,
+      "first_token_acc": 0.34375,
+      "token_acc": 0.045572916666666664,
       "exact_seq_acc": 0.0
     },
     {
       "bake_steps": 16,
-      "prompt_overlap": 10,
-      "first_token_acc": 0.0,
-      "token_acc": 0.026041666666666668,
+      "prompt_overlap": 16,
+      "first_token_acc": 0.09375,
+      "token_acc": 0.02734375,
       "exact_seq_acc": 0.0
     },
     {
-      "bake_steps": 64,
+      "bake_steps": 96,
       "prompt_overlap": 0,
-      "first_token_acc": 0.65625,
-      "token_acc": 0.15104166666666666,
+      "first_token_acc": 0.59375,
+      "token_acc": 0.10677083333333333,
       "exact_seq_acc": 0.0
     },
     {
-      "bake_steps": 64,
-      "prompt_overlap": 10,
+      "bake_steps": 96,
+      "prompt_overlap": 16,
       "first_token_acc": 0.6875,
-      "token_acc": 0.15755208333333334,
+      "token_acc": 0.12239583333333333,
       "exact_seq_acc": 0.0
     }
   ]