Spaces:
Running
Running
| <div> <div id="6f634a12-ce30-4ac5-b729-53fa56977873" class="plotly-graph-div" style="height:100%; width:100%;"></div> <script type="text/javascript"> window.PLOTLYENV=window.PLOTLYENV || {}; if (document.getElementById("6f634a12-ce30-4ac5-b729-53fa56977873")) { Plotly.newPlot( "6f634a12-ce30-4ac5-b729-53fa56977873", [{"hoverinfo":"all","marker":{"color":"#FF323D"},"mode":"markers","name":"MMLU-PRO as function of (MMLU)","text":["01-ai\u002fYi-1.5-34B","01-ai\u002fYi-1.5-34B-32K","01-ai\u002fYi-1.5-34B-Chat","01-ai\u002fYi-1.5-6B","01-ai\u002fYi-1.5-6B","01-ai\u002fYi-1.5-6B-Chat","01-ai\u002fYi-1.5-9B","01-ai\u002fYi-1.5-9B-32K","01-ai\u002fYi-1.5-9B-Chat","01-ai\u002fYi-1.5-9B-Chat-16K","01-ai\u002fYi-34B","01-ai\u002fYi-34B-200K","01-ai\u002fYi-34B-Chat","01-ai\u002fYi-34B-Chat","01-ai\u002fYi-6B","01-ai\u002fYi-6B","01-ai\u002fYi-6B-200K","01-ai\u002fYi-6B-200K","01-ai\u002fYi-9B","01-ai\u002fYi-9B-200K","CausalLM\u002f14B","CohereForAI\u002faya-23-8B","CohereForAI\u002fc4ai-command-r-plus","CohereForAI\u002fc4ai-command-r-plus","CohereForAI\u002fc4ai-command-r-v01","Deci\u002fDeciLM-7B","Deci\u002fDeciLM-7B-instruct","EleutherAI\u002fgpt-j-6b","EleutherAI\u002fgpt-neo-1.3B","EleutherAI\u002fgpt-neo-2.7B","EleutherAI\u002fgpt-neox-20b","EleutherAI\u002fpythia-12b","EleutherAI\u002fpythia-160m","EleutherAI\u002fpythia-410m","HuggingFaceH4\u002fzephyr-7b-alpha","HuggingFaceH4\u002fzephyr-7b-beta","HuggingFaceH4\u002fzephyr-7b-beta","HuggingFaceH4\u002fzephyr-7b-beta","HuggingFaceH4\u002fzephyr-7b-gemma-v0.1","HuggingFaceH4\u002fzephyr-7b-gemma-v0.1","IDEA-CCNL\u002fZiya-LLaMA-13B-v1","Intel\u002fneural-chat-7b-v3","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-2","Intel\u002fneural-chat-7b-v3-3","Nexusflow\u002fNexusRaven-V2-13B","NousResearch\u002fHermes-2-Pro-Llama-3-8B","NousResearch\u002fHermes-2-Pro-Mistral-7B","NousResearch\u002fHermes-2-Pro-Mistral-7B","NousResearch\u002fHermes-2-Theta-Llama-3-8B","NousResearch\u002fHermes-2-Theta-Llama-3-8B","NousResearch\u002fNous-Hermes-2-Mistral-7B-DPO","NousResearch\u002fNous-Hermes-2-Mistral-7B-DPO","NousResearch\u002fNous-Hermes-2-Mixtral-8x7B-DPO","NousResearch\u002fNous-Hermes-2-Mixtral-8x7B-DPO","NousResearch\u002fNous-Hermes-2-Mixtral-8x7B-SFT","NousResearch\u002fNous-Hermes-2-SOLAR-10.7B","NousResearch\u002fNous-Hermes-llama-2-7b","NousResearch\u002fYarn-Mistral-7b-128k","NousResearch\u002fYarn-Mistral-7b-64k","Open-Orca\u002fMistral-7B-OpenOrca","OpenAssistant\u002foasst-sft-1-pythia-12b","PygmalionAI\u002fpygmalion-6b","Qwen\u002fQwen1.5-0.5B","Qwen\u002fQwen1.5-0.5B-Chat","Qwen\u002fQwen1.5-1.8B","Qwen\u002fQwen1.5-1.8B-Chat","Qwen\u002fQwen1.5-110B","Qwen\u002fQwen1.5-110B-Chat","Qwen\u002fQwen1.5-14B","Qwen\u002fQwen1.5-14B-Chat","Qwen\u002fQwen1.5-14B-Chat","Qwen\u002fQwen1.5-32B","Qwen\u002fQwen1.5-32B","Qwen\u002fQwen1.5-32B-Chat","Qwen\u002fQwen1.5-4B","Qwen\u002fQwen1.5-4B-Chat","Qwen\u002fQwen1.5-7B","Qwen\u002fQwen1.5-7B-Chat","Qwen\u002fQwen1.5-7B-Chat","Qwen\u002fQwen1.5-MoE-A2.7B","Qwen\u002fQwen1.5-MoE-A2.7B-Chat","Qwen\u002fQwen2-0.5B","Qwen\u002fQwen2-1.5B","Qwen\u002fQwen2-7B","TencentARC\u002fLLaMA-Pro-8B","TencentARC\u002fLLaMA-Pro-8B-Instruct","TencentARC\u002fMistral_Pro_8B_v0.1","abacusai\u002fSmaug-34B-v0.1","abacusai\u002fSmaug-72B-v0.1","allenai\u002fOLMo-1B-hf","allenai\u002fOLMo-1B-hf","argilla\u002fnotux-8x7b-v1","berkeley-nest\u002fStarling-LM-7B-alpha","berkeley-nest\u002fStarling-LM-7B-alpha","bigcode\u002fstarcoder2-15b","bigcode\u002fstarcoder2-3b","bigcode\u002fstarcoder2-7b","bigscience\u002fbloom-1b1","bigscience\u002fbloom-1b7","bigscience\u002fbloom-3b","bigscience\u002fbloom-560m","bigscience\u002fbloom-7b1","cognitivecomputations\u002fdolphin-2.9-llama3-8b","cognitivecomputations\u002fdolphin-2.9-llama3-8b","databricks\u002fdolly-v2-12b","databricks\u002fdolly-v2-3b","databricks\u002fdolly-v2-7b","deepseek-ai\u002fdeepseek-llm-67b-chat","deepseek-ai\u002fdeepseek-llm-67b-chat","deepseek-ai\u002fdeepseek-llm-7b-chat","deepseek-ai\u002fdeepseek-llm-7b-chat","deepseek-ai\u002fdeepseek-moe-16b-base","facebook\u002fopt-1.3b","facebook\u002fopt-30b","google\u002fgemma-1.1-7b-it","google\u002fgemma-2b","google\u002fgemma-2b","google\u002fgemma-2b-it","google\u002fgemma-7b","google\u002fgemma-7b","google\u002fgemma-7b-it","google\u002frecurrentgemma-2b","google\u002frecurrentgemma-2b","google\u002frecurrentgemma-2b-it","gpt2","gpt2","gradientai\u002fLlama-3-8B-Instruct-Gradient-1048k","gradientai\u002fLlama-3-8B-Instruct-Gradient-1048k","ibm\u002fmerlinite-7b","lmsys\u002fvicuna-7b-v1.5","lmsys\u002fvicuna-7b-v1.5","meta-llama\u002fLlama-2-13b-chat-hf","meta-llama\u002fLlama-2-13b-hf","meta-llama\u002fLlama-2-70b-chat-hf","meta-llama\u002fLlama-2-70b-hf","meta-llama\u002fLlama-2-7b-chat-hf","meta-llama\u002fLlama-2-7b-hf","meta-llama\u002fLlama-2-7b-hf","meta-llama\u002fMeta-Llama-3-70B","meta-llama\u002fMeta-Llama-3-70B-Instruct","meta-llama\u002fMeta-Llama-3-8B","meta-llama\u002fMeta-Llama-3-8B","meta-llama\u002fMeta-Llama-3-8B-Instruct","microsoft\u002fDialoGPT-medium","microsoft\u002fOrca-2-13b","microsoft\u002fOrca-2-13b","microsoft\u002fOrca-2-7b","microsoft\u002fPhi-3-medium-4k-instruct","microsoft\u002fPhi-3-mini-128k-instruct","microsoft\u002fPhi-3-mini-4k-instruct","microsoft\u002fphi-1_5","microsoft\u002fphi-2","microsoft\u002fphi-2","mistral-community\u002fMistral-7B-v0.2","mistral-community\u002fMistral-7B-v0.2","mistralai\u002fMistral-7B-Instruct-v0.1","mistralai\u002fMistral-7B-Instruct-v0.2","mistralai\u002fMistral-7B-v0.1","mistralai\u002fMistral-7B-v0.3","mistralai\u002fMixtral-8x7B-Instruct-v0.1","mistralai\u002fMixtral-8x7B-Instruct-v0.1","mistralai\u002fMixtral-8x7B-v0.1","mistralai\u002fMixtral-8x7B-v0.1","mistralai\u002fMixtral-8x7B-v0.1","mlabonne\u002fAlphaMonarch-7B","mlabonne\u002fBeyonder-4x7B-v3","mlabonne\u002fOrpoLlama-3-8B","mosaicml\u002fmpt-7b","openai-community\u002fgpt2","openai-community\u002fgpt2-large","openchat\u002fopenchat-3.5-1210","openchat\u002fopenchat_3.5","openchat\u002fopenchat_3.5","openchat\u002fopenchat_3.5","openchat\u002fopenchat_v3.2","openchat\u002fopenchat_v3.2","openchat\u002fopenchat_v3.2_super","refuelai\u002fLlama-3-Refueled","refuelai\u002fLlama-3-Refueled","stabilityai\u002fstablelm-2-12b","stabilityai\u002fstablelm-2-12b-chat","stabilityai\u002fstablelm-2-1_6b","stabilityai\u002fstablelm-2-1_6b-chat","stabilityai\u002fstablelm-2-zephyr-1_6b","stabilityai\u002fstablelm-3b-4e1t","stabilityai\u002fstablelm-zephyr-3b","teknium\u002fCollectiveCognition-v1.1-Mistral-7B","teknium\u002fOpenHermes-13B","teknium\u002fOpenHermes-2.5-Mistral-7B","teknium\u002fOpenHermes-2.5-Mistral-7B","teknium\u002fOpenHermes-7B","tiiuae\u002ffalcon-11B","tiiuae\u002ffalcon-40b","tiiuae\u002ffalcon-7b","tiiuae\u002ffalcon-7b-instruct","tiiuae\u002ffalcon-7b-instruct","togethercomputer\u002fGPT-JT-6B-v1","togethercomputer\u002fGPT-NeoXT-Chat-Base-20B","togethercomputer\u002fLLaMA-2-7B-32K","togethercomputer\u002fLlama-2-7B-32K-Instruct","togethercomputer\u002fLlama-2-7B-32K-Instruct","togethercomputer\u002fRedPajama-INCITE-7B-Base","togethercomputer\u002fRedPajama-INCITE-7B-Chat","togethercomputer\u002fRedPajama-INCITE-7B-Instruct","togethercomputer\u002fRedPajama-INCITE-Base-3B-v1","togethercomputer\u002fRedPajama-INCITE-Chat-3B-v1","togethercomputer\u002fRedPajama-INCITE-Instruct-3B-v1","upstage\u002fSOLAR-10.7B-Instruct-v1.0","upstage\u002fSOLAR-10.7B-v1.0"],"x":[77.9957192730956,78.15303198667405,77.08283992203879,64.72689535017567,65.00272042688377,64.24095801586486,70.73451954519469,70.97203740649417,71.01349610420871,70.95946328062593,76.35156727391681,75.56180045111563,74.86694971720243,74.89765262226099,63.84606128348609,64.10709139827398,64.64867451074461,64.64930135205712,70.06330822380845,70.34013960436471,65.86320847049085,56.33899126009816,75.72824616707308,74.02494713760785,68.19530224441294,59.762938924439105,60.24359775384097,26.783999358640752,24.821655499137048,26.45397884474629,25.002595295235746,26.75606660161476,24.950593393502686,27.245507410753593,61.3896585162216,61.061389920377685,61.06911135758283,60.70490193884751,58.975890703009014,60.682678692224144,27.03799242337422,62.26105246211977,60.791623329684086,62.374203705324284,62.44460409845261,62.11841694257893,63.550105055987174,63.07073351412039,44.87880425591219,64.79850156958106,62.20609448217756,62.1246057414326,66.51191133673278,66.55303373586858,63.356659125507434,63.245294478077106,72.27737838264045,72.17054547051251,72.20739527153786,66.29541259257003,48.335234653692204,63.02324877855956,62.96478981633615,62.24441734951871,26.193572189883703,25.730836314001447,39.35334081217581,33.81983654590648,46.70838530175783,45.86651800528403,80.19575908781562,78.03946337687869,69.36026470831474,68.51540963742676,68.57472555756478,74.29961283965316,74.31221659340622,74.99313230796709,56.523519639933376,55.55462512266324,61.9700216814056,61.64766854906246,61.69764020685532,62.54071682523817,60.970670017639726,45.17037929235293,57.25919892204937,70.4515868286805,47.49331993287192,52.577029472650906,61.74163205184785,76.66476969478123,77.15452603998116,26.234231797093017,26.306692338086823,71.38894203721195,64.70438661462165,64.66799008434704,51.34583865906561,38.64860206508296,41.21331864479665,26.70214825620786,27.47846514752041,26.592509199327857,24.21541212062042,26.246222375725694,61.03678038351008,61.04002830421494,25.916845651522607,24.696321332710696,25.183937022975865,72.41916167903139,72.18744401223967,51.74656618684998,51.77213745416527,46.306113561052754,24.963045825616692,26.658730035291097,60.91642429212937,41.679249280698635,41.77314644002599,37.648186828291585,64.5598699571695,66.02745725247713,53.5170310018511,34.611140710007234,34.38230443417011,40.86948703219191,25.833717804860335,25.404197294686202,61.917687321536896,62.84956375246854,64.90605869704184,51.037449016607304,50.8231434865938,54.63627087850933,55.77103633323489,63.90701952816292,69.83208921288625,48.318825530104384,46.866075004498,43.79609636092417,79.23293386179718,80.05922317995427,66.70480406193239,66.49495492503577,67.0721995784583,25.838967331895883,60.37069885710225,60.301159128932504,56.371153419443566,77.83091143434687,68.70399293043886,69.08301379498349,43.886467784197876,58.1123279817493,57.92026667108576,63.689429588238426,63.617802578997775,55.37547435463056,60.778011275770005,64.16448378492244,63.46289361741425,71.16228271617086,71.39615714693673,71.81543655466032,70.17346542519046,71.87839332236227,64.40157643349518,64.87016580791871,65.76439370548329,30.795325425385716,25.833717804860335,26.067403063531568,64.62163338646481,64.97554013321405,64.94207335284045,62.892016034379076,56.823746005051746,56.68268469601176,55.89931756667269,64.73176939664393,64.68760308408112,62.039927374602314,61.05866061151978,38.94657259915192,41.47262454451639,42.03478315758355,45.225737544229936,46.1679465865397,62.349785875539155,56.34727779629169,63.8216682847302,63.636962953801635,48.617757574363246,58.3670122238434,56.8938511726681,27.785470153285935,25.660046534596926,25.836963794035032,47.19018110516562,29.919101184126255,43.32570701968771,46.105045377255685,45.53223866033279,27.68214018695477,26.942822650634863,37.61836019303777,27.02787422464126,26.231263349013634,25.032214024865144,66.21064893071143,65.47877433224612],"y":[40.732121749408975,41.21232269503546,39.11606087470449,23.343306737588648,23.343306737588648,24.414524231678485,32.402482269503544,30.721778959810877,33.05814125295508,33.261303191489354,37.90632387706855,39.273049645390074,34.369459219858165,34.369459219858165,22.12433510638298,22.12433510638298,20.489804964539008,20.489804964539008,28.59781323877068,29.1334219858156,24.682328605200944,14.2010195035461,33.242833924349874,33.242833924349874,26.326093380614658,18.799867021276594,17.86716903073286,2.6761968085106376,1.8173758865248217,1.8081412529550822,1.725029550827422,1.2078900709219857,1.3279403073286051,1.4202866430260035,19.94496158392435,19.78797281323877,19.78797281323877,19.78797281323877,20.526743498817968,20.526743498817968,1.124778368794326,18.873744089834513,18.642878250591018,18.642878250591018,18.642878250591018,18.642878250591018,18.522828014184395,18.05186170212766,9.685283687943262,22.798463356973993,21.625664893617017,21.625664893617017,26.316858747044915,26.316858747044915,22.392139479905435,22.392139479905435,29.63209219858156,29.63209219858156,22.95545212765957,27.314199172576835,10.442523640661937,21.034648345153663,21.265514184397162,18.36583924349882,1.2540632387706852,2.0390070921985806,3.4149674940898342,2.362219267139479,9.79609929078014,8.928043735224584,48.45227541371159,42.495936761229316,29.373522458628837,29.08724881796691,29.08724881796691,38.88519503546098,38.88519503546098,38.41422872340425,16.22340425531915,15.512337470449172,21.293218085106382,21.681072695035464,21.681072695035464,19.751034278959807,21.3670951536643,7.7552452718676115,17.239213947990542,35.366799645390074,9.011155437352246,10.507166075650117,19.61251477541371,39.36539598108747,40.45508274231678,1.9281914893617011,1.9281914893617011,29.659796099290777,24.128250591016545,24.128250591016545,15.032136524822693,7.0718823877068555,7.1365248226950335,1.1986554373522447,0.9585549645390061,1.4756944444444438,1.8266105200945615,1.1617169030732852,19.67715721040189,19.67715721040189,1.4295212765957446,1.6142139479905429,1.6603871158392434,32.70722517730496,32.70722517730496,12.594193262411347,12.594193262411347,5.612810283687943,1.1894208037825047,1.8173758865248217,17.5993646572104,4.061391843971631,4.061391843971631,3.9228723404255303,21.644134160756497,21.644134160756497,7.7183067375886525,1.9928339243498814,1.9928339243498814,4.4123079196217505,1.7712027186761226,1.7712027186761226,21.56102245862884,21.56102245862884,22.983156028368796,12.741947399527188,12.741947399527188,10.257830969267138,15.309175531914892,15.918661347517732,30.1954048463357,7.524379432624113,9.56523345153664,9.56523345153664,41.21232269503546,46.74386820330969,24.553043735224584,24.553043735224584,29.604388297872337,1.3187056737588652,19.437056737588648,19.437056737588648,14.653516548463358,40.84293735224587,30.38009751773049,31.848404255319146,7.6813682033096935,18.088800236406616,18.088800236406616,21.699541962174944,21.699541962174944,15.336879432624112,19.076906028368796,22.364435579196215,21.699541962174944,29.364287825059105,29.364287825059105,31.49748817966903,31.49748817966903,31.49748817966903,16.36192375886525,16.805186170212764,18.947621158392433,2.288342198581559,1.7712027186761226,1.5772754137115832,23.805038416075647,23.92508865248227,23.92508865248227,23.92508865248227,15.79861111111111,15.79861111111111,15.83554964539007,23.278664302600472,23.278664302600472,23.020094562647756,19.270833333333332,5.1510786052009445,6.905658983451536,7.930703309692672,7.432033096926712,8.530954491725769,20.40669326241135,15.438460401891252,22.826167257683213,22.826167257683213,10.368646572104018,15.438460401891252,16.722074468085104,1.392582742316784,1.725029550827422,1.725029550827422,6.951832151300234,1.6142139479905429,8.530954491725769,8.678708628841607,8.678708628841607,2.186761229314421,1.3464095744680846,3.027112884160755,1.2355939716312052,1.4110520094562635,1.2171247044917257,23.758865248226947,26.667774822695034],"type":"scatter"},{"hoverinfo":"all","marker":{"color":"#FF9D00"},"mode":"markers","name":"GPQA as function of (MMLU)","text":["01-ai\u002fYi-1.5-34B","01-ai\u002fYi-1.5-34B-32K","01-ai\u002fYi-1.5-34B-Chat","01-ai\u002fYi-1.5-6B","01-ai\u002fYi-1.5-6B","01-ai\u002fYi-1.5-6B-Chat","01-ai\u002fYi-1.5-9B","01-ai\u002fYi-1.5-9B-32K","01-ai\u002fYi-1.5-9B-Chat","01-ai\u002fYi-1.5-9B-Chat-16K","01-ai\u002fYi-34B","01-ai\u002fYi-34B-200K","01-ai\u002fYi-34B-Chat","01-ai\u002fYi-34B-Chat","01-ai\u002fYi-6B","01-ai\u002fYi-6B","01-ai\u002fYi-6B-200K","01-ai\u002fYi-6B-200K","01-ai\u002fYi-9B","01-ai\u002fYi-9B-200K","CausalLM\u002f14B","CohereForAI\u002faya-23-8B","CohereForAI\u002fc4ai-command-r-plus","CohereForAI\u002fc4ai-command-r-plus","CohereForAI\u002fc4ai-command-r-v01","Deci\u002fDeciLM-7B","Deci\u002fDeciLM-7B-instruct","EleutherAI\u002fgpt-j-6b","EleutherAI\u002fgpt-neo-1.3B","EleutherAI\u002fgpt-neo-2.7B","EleutherAI\u002fgpt-neox-20b","EleutherAI\u002fpythia-12b","EleutherAI\u002fpythia-160m","EleutherAI\u002fpythia-410m","HuggingFaceH4\u002fzephyr-7b-alpha","HuggingFaceH4\u002fzephyr-7b-beta","HuggingFaceH4\u002fzephyr-7b-beta","HuggingFaceH4\u002fzephyr-7b-beta","HuggingFaceH4\u002fzephyr-7b-gemma-v0.1","HuggingFaceH4\u002fzephyr-7b-gemma-v0.1","IDEA-CCNL\u002fZiya-LLaMA-13B-v1","Intel\u002fneural-chat-7b-v3","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-1","Intel\u002fneural-chat-7b-v3-2","Intel\u002fneural-chat-7b-v3-3","Nexusflow\u002fNexusRaven-V2-13B","NousResearch\u002fHermes-2-Pro-Llama-3-8B","NousResearch\u002fHermes-2-Pro-Mistral-7B","NousResearch\u002fHermes-2-Pro-Mistral-7B","NousResearch\u002fHermes-2-Theta-Llama-3-8B","NousResearch\u002fHermes-2-Theta-Llama-3-8B","NousResearch\u002fNous-Hermes-2-Mistral-7B-DPO","NousResearch\u002fNous-Hermes-2-Mistral-7B-DPO","NousResearch\u002fNous-Hermes-2-Mixtral-8x7B-DPO","NousResearch\u002fNous-Hermes-2-Mixtral-8x7B-DPO","NousResearch\u002fNous-Hermes-2-Mixtral-8x7B-SFT","NousResearch\u002fNous-Hermes-2-SOLAR-10.7B","NousResearch\u002fNous-Hermes-llama-2-7b","NousResearch\u002fYarn-Mistral-7b-128k","NousResearch\u002fYarn-Mistral-7b-64k","Open-Orca\u002fMistral-7B-OpenOrca","OpenAssistant\u002foasst-sft-1-pythia-12b","PygmalionAI\u002fpygmalion-6b","Qwen\u002fQwen1.5-0.5B","Qwen\u002fQwen1.5-0.5B-Chat","Qwen\u002fQwen1.5-1.8B","Qwen\u002fQwen1.5-1.8B-Chat","Qwen\u002fQwen1.5-110B","Qwen\u002fQwen1.5-110B-Chat","Qwen\u002fQwen1.5-14B","Qwen\u002fQwen1.5-14B-Chat","Qwen\u002fQwen1.5-14B-Chat","Qwen\u002fQwen1.5-32B","Qwen\u002fQwen1.5-32B","Qwen\u002fQwen1.5-32B-Chat","Qwen\u002fQwen1.5-4B","Qwen\u002fQwen1.5-4B-Chat","Qwen\u002fQwen1.5-7B","Qwen\u002fQwen1.5-7B-Chat","Qwen\u002fQwen1.5-7B-Chat","Qwen\u002fQwen1.5-MoE-A2.7B","Qwen\u002fQwen1.5-MoE-A2.7B-Chat","Qwen\u002fQwen2-0.5B","Qwen\u002fQwen2-1.5B","Qwen\u002fQwen2-7B","TencentARC\u002fLLaMA-Pro-8B","TencentARC\u002fLLaMA-Pro-8B-Instruct","TencentARC\u002fMistral_Pro_8B_v0.1","abacusai\u002fSmaug-34B-v0.1","abacusai\u002fSmaug-72B-v0.1","allenai\u002fOLMo-1B-hf","allenai\u002fOLMo-1B-hf","argilla\u002fnotux-8x7b-v1","berkeley-nest\u002fStarling-LM-7B-alpha","berkeley-nest\u002fStarling-LM-7B-alpha","bigcode\u002fstarcoder2-15b","bigcode\u002fstarcoder2-3b","bigcode\u002fstarcoder2-7b","bigscience\u002fbloom-1b1","bigscience\u002fbloom-1b7","bigscience\u002fbloom-3b","bigscience\u002fbloom-560m","bigscience\u002fbloom-7b1","cognitivecomputations\u002fdolphin-2.9-llama3-8b","cognitivecomputations\u002fdolphin-2.9-llama3-8b","databricks\u002fdolly-v2-12b","databricks\u002fdolly-v2-3b","databricks\u002fdolly-v2-7b","deepseek-ai\u002fdeepseek-llm-67b-chat","deepseek-ai\u002fdeepseek-llm-67b-chat","deepseek-ai\u002fdeepseek-llm-7b-chat","deepseek-ai\u002fdeepseek-llm-7b-chat","deepseek-ai\u002fdeepseek-moe-16b-base","facebook\u002fopt-1.3b","facebook\u002fopt-30b","google\u002fgemma-1.1-7b-it","google\u002fgemma-2b","google\u002fgemma-2b","google\u002fgemma-2b-it","google\u002fgemma-7b","google\u002fgemma-7b","google\u002fgemma-7b-it","google\u002frecurrentgemma-2b","google\u002frecurrentgemma-2b","google\u002frecurrentgemma-2b-it","gpt2","gpt2","gradientai\u002fLlama-3-8B-Instruct-Gradient-1048k","gradientai\u002fLlama-3-8B-Instruct-Gradient-1048k","ibm\u002fmerlinite-7b","lmsys\u002fvicuna-7b-v1.5","lmsys\u002fvicuna-7b-v1.5","meta-llama\u002fLlama-2-13b-chat-hf","meta-llama\u002fLlama-2-13b-hf","meta-llama\u002fLlama-2-70b-chat-hf","meta-llama\u002fLlama-2-70b-hf","meta-llama\u002fLlama-2-7b-chat-hf","meta-llama\u002fLlama-2-7b-hf","meta-llama\u002fLlama-2-7b-hf","meta-llama\u002fMeta-Llama-3-70B","meta-llama\u002fMeta-Llama-3-70B-Instruct","meta-llama\u002fMeta-Llama-3-8B","meta-llama\u002fMeta-Llama-3-8B","meta-llama\u002fMeta-Llama-3-8B-Instruct","microsoft\u002fDialoGPT-medium","microsoft\u002fOrca-2-13b","microsoft\u002fOrca-2-13b","microsoft\u002fOrca-2-7b","microsoft\u002fPhi-3-medium-4k-instruct","microsoft\u002fPhi-3-mini-128k-instruct","microsoft\u002fPhi-3-mini-4k-instruct","microsoft\u002fphi-1_5","microsoft\u002fphi-2","microsoft\u002fphi-2","mistral-community\u002fMistral-7B-v0.2","mistral-community\u002fMistral-7B-v0.2","mistralai\u002fMistral-7B-Instruct-v0.1","mistralai\u002fMistral-7B-Instruct-v0.2","mistralai\u002fMistral-7B-v0.1","mistralai\u002fMistral-7B-v0.3","mistralai\u002fMixtral-8x7B-Instruct-v0.1","mistralai\u002fMixtral-8x7B-Instruct-v0.1","mistralai\u002fMixtral-8x7B-v0.1","mistralai\u002fMixtral-8x7B-v0.1","mistralai\u002fMixtral-8x7B-v0.1","mlabonne\u002fAlphaMonarch-7B","mlabonne\u002fBeyonder-4x7B-v3","mlabonne\u002fOrpoLlama-3-8B","mosaicml\u002fmpt-7b","openai-community\u002fgpt2","openai-community\u002fgpt2-large","openchat\u002fopenchat-3.5-1210","openchat\u002fopenchat_3.5","openchat\u002fopenchat_3.5","openchat\u002fopenchat_3.5","openchat\u002fopenchat_v3.2","openchat\u002fopenchat_v3.2","openchat\u002fopenchat_v3.2_super","refuelai\u002fLlama-3-Refueled","refuelai\u002fLlama-3-Refueled","stabilityai\u002fstablelm-2-12b","stabilityai\u002fstablelm-2-12b-chat","stabilityai\u002fstablelm-2-1_6b","stabilityai\u002fstablelm-2-1_6b-chat","stabilityai\u002fstablelm-2-zephyr-1_6b","stabilityai\u002fstablelm-3b-4e1t","stabilityai\u002fstablelm-zephyr-3b","teknium\u002fCollectiveCognition-v1.1-Mistral-7B","teknium\u002fOpenHermes-13B","teknium\u002fOpenHermes-2.5-Mistral-7B","teknium\u002fOpenHermes-2.5-Mistral-7B","teknium\u002fOpenHermes-7B","tiiuae\u002ffalcon-11B","tiiuae\u002ffalcon-40b","tiiuae\u002ffalcon-7b","tiiuae\u002ffalcon-7b-instruct","tiiuae\u002ffalcon-7b-instruct","togethercomputer\u002fGPT-JT-6B-v1","togethercomputer\u002fGPT-NeoXT-Chat-Base-20B","togethercomputer\u002fLLaMA-2-7B-32K","togethercomputer\u002fLlama-2-7B-32K-Instruct","togethercomputer\u002fLlama-2-7B-32K-Instruct","togethercomputer\u002fRedPajama-INCITE-7B-Base","togethercomputer\u002fRedPajama-INCITE-7B-Chat","togethercomputer\u002fRedPajama-INCITE-7B-Instruct","togethercomputer\u002fRedPajama-INCITE-Base-3B-v1","togethercomputer\u002fRedPajama-INCITE-Chat-3B-v1","togethercomputer\u002fRedPajama-INCITE-Instruct-3B-v1","upstage\u002fSOLAR-10.7B-Instruct-v1.0","upstage\u002fSOLAR-10.7B-v1.0"],"x":[77.9957192730956,78.15303198667405,77.08283992203879,64.72689535017567,65.00272042688377,64.24095801586486,70.73451954519469,70.97203740649417,71.01349610420871,70.95946328062593,76.35156727391681,75.56180045111563,74.86694971720243,74.89765262226099,63.84606128348609,64.10709139827398,64.64867451074461,64.64930135205712,70.06330822380845,70.34013960436471,65.86320847049085,56.33899126009816,75.72824616707308,74.02494713760785,68.19530224441294,59.762938924439105,60.24359775384097,26.783999358640752,24.821655499137048,26.45397884474629,25.002595295235746,26.75606660161476,24.950593393502686,27.245507410753593,61.3896585162216,61.061389920377685,61.06911135758283,60.70490193884751,58.975890703009014,60.682678692224144,27.03799242337422,62.26105246211977,60.791623329684086,62.374203705324284,62.44460409845261,62.11841694257893,63.550105055987174,63.07073351412039,44.87880425591219,64.79850156958106,62.20609448217756,62.1246057414326,66.51191133673278,66.55303373586858,63.356659125507434,63.245294478077106,72.27737838264045,72.17054547051251,72.20739527153786,66.29541259257003,48.335234653692204,63.02324877855956,62.96478981633615,62.24441734951871,26.193572189883703,25.730836314001447,39.35334081217581,33.81983654590648,46.70838530175783,45.86651800528403,80.19575908781562,78.03946337687869,69.36026470831474,68.51540963742676,68.57472555756478,74.29961283965316,74.31221659340622,74.99313230796709,56.523519639933376,55.55462512266324,61.9700216814056,61.64766854906246,61.69764020685532,62.54071682523817,60.970670017639726,45.17037929235293,57.25919892204937,70.4515868286805,47.49331993287192,52.577029472650906,61.74163205184785,76.66476969478123,77.15452603998116,26.234231797093017,26.306692338086823,71.38894203721195,64.70438661462165,64.66799008434704,51.34583865906561,38.64860206508296,41.21331864479665,26.70214825620786,27.47846514752041,26.592509199327857,24.21541212062042,26.246222375725694,61.03678038351008,61.04002830421494,25.916845651522607,24.696321332710696,25.183937022975865,72.41916167903139,72.18744401223967,51.74656618684998,51.77213745416527,46.306113561052754,24.963045825616692,26.658730035291097,60.91642429212937,41.679249280698635,41.77314644002599,37.648186828291585,64.5598699571695,66.02745725247713,53.5170310018511,34.611140710007234,34.38230443417011,40.86948703219191,25.833717804860335,25.404197294686202,61.917687321536896,62.84956375246854,64.90605869704184,51.037449016607304,50.8231434865938,54.63627087850933,55.77103633323489,63.90701952816292,69.83208921288625,48.318825530104384,46.866075004498,43.79609636092417,79.23293386179718,80.05922317995427,66.70480406193239,66.49495492503577,67.0721995784583,25.838967331895883,60.37069885710225,60.301159128932504,56.371153419443566,77.83091143434687,68.70399293043886,69.08301379498349,43.886467784197876,58.1123279817493,57.92026667108576,63.689429588238426,63.617802578997775,55.37547435463056,60.778011275770005,64.16448378492244,63.46289361741425,71.16228271617086,71.39615714693673,71.81543655466032,70.17346542519046,71.87839332236227,64.40157643349518,64.87016580791871,65.76439370548329,30.795325425385716,25.833717804860335,26.067403063531568,64.62163338646481,64.97554013321405,64.94207335284045,62.892016034379076,56.823746005051746,56.68268469601176,55.89931756667269,64.73176939664393,64.68760308408112,62.039927374602314,61.05866061151978,38.94657259915192,41.47262454451639,42.03478315758355,45.225737544229936,46.1679465865397,62.349785875539155,56.34727779629169,63.8216682847302,63.636962953801635,48.617757574363246,58.3670122238434,56.8938511726681,27.785470153285935,25.660046534596926,25.836963794035032,47.19018110516562,29.919101184126255,43.32570701968771,46.105045377255685,45.53223866033279,27.68214018695477,26.942822650634863,37.61836019303777,27.02787422464126,26.231263349013634,25.032214024865144,66.21064893071143,65.47877433224612],"y":[15.436241610738257,15.100671140939594,15.324384787472036,8.277404921700223,8.277404921700223,9.060402684563762,17.225950782997764,14.541387024608499,11.297539149888143,7.829977628635347,15.548098434004473,14.205816554809845,11.74496644295302,11.74496644295302,2.572706935123044,2.572706935123044,4.250559284116329,4.250559284116329,9.060402684563762,8.7248322147651,7.046979865771815,4.5861297539149914,7.38255033557047,7.38255033557047,7.606263982102905,6.040268456375841,5.257270693512303,0.0,0.7829977628635317,2.1252796420581683,0.0,0.0,1.1185682326621946,1.230425055928408,6.375838926174497,5.369127516778524,5.369127516778524,5.369127516778524,5.92841163310962,5.92841163310962,0.0,5.592841163310966,5.369127516778524,5.369127516778524,5.369127516778524,5.369127516778524,5.369127516778524,5.257270693512303,1.342281879194629,5.7046979865771785,3.1319910514541416,3.1319910514541416,7.158836689038028,7.158836689038028,5.7046979865771785,5.7046979865771785,9.61968680089485,9.61968680089485,6.935123042505594,5.8165548098433995,1.7897091722595053,6.487695749440718,5.369127516778524,2.9082774049216997,1.0067114093959737,0.0,0.5592841163310973,2.572706935123044,7.38255033557047,6.375838926174497,13.646532438478745,12.192393736017896,5.92841163310962,2.684563758389265,2.684563758389265,10.626398210290827,10.626398210290827,7.494407158836691,3.5794183445190177,2.2371364653243813,6.487695749440718,7.046979865771815,7.046979865771815,1.230425055928408,3.243847874720355,0.7829977628635317,1.9015659955257262,7.270693512304249,1.342281879194629,3.243847874720355,4.026845637583895,10.626398210290827,9.61968680089485,1.5659955257270708,1.5659955257270708,7.494407158836691,6.263982102908276,6.263982102908276,2.9082774049216997,0.0,0.22371364653244186,1.230425055928408,1.1185682326621946,0.0,1.5659955257270708,1.9015659955257262,4.921700223713646,4.921700223713646,0.0,1.4541387024608499,2.460850111856823,8.83668903803132,8.83668903803132,2.1252796420581683,2.1252796420581683,0.5592841163310973,0.0,2.572706935123044,5.8165548098433995,0.6711409395973182,0.6711409395973182,3.8031319910514525,4.921700223713646,4.921700223713646,4.5861297539149914,0.0,0.0,1.0067114093959737,1.1185682326621946,1.1185682326621946,3.6912751677852316,3.6912751677852316,6.263982102908276,1.1185682326621946,1.1185682326621946,0.0,4.138702460850116,1.9015659955257262,7.046979865771815,0.5592841163310973,2.2371364653243813,2.2371364653243813,19.686800894854585,4.921700223713646,7.38255033557047,7.38255033557047,1.230425055928408,0.5592841163310973,4.026845637583895,4.026845637583895,0.7829977628635317,11.521252796420578,9.060402684563762,9.284116331096197,2.348993288590602,2.9082774049216997,2.9082774049216997,5.592841163310966,5.592841163310966,0.0,3.467561521252797,5.592841163310966,5.592841163310966,7.606263982102905,7.606263982102905,9.284116331096197,9.284116331096197,9.284116331096197,2.684563758389265,4.697986577181204,3.9149888143176734,1.342281879194629,1.1185682326621946,1.230425055928408,6.823266219239373,6.487695749440718,6.487695749440718,6.487695749440718,2.684563758389265,2.684563758389265,1.9015659955257262,6.599552572706939,6.599552572706939,3.8031319910514525,2.2371364653243813,0.0,0.0,0.0,0.0,0.0,4.921700223713646,3.0201342281879207,4.4742729306487705,4.4742729306487705,2.572706935123044,2.796420581655479,3.1319910514541416,0.0,0.0,0.0,1.4541387024608499,0.0,0.0,0.22371364653244186,0.22371364653244186,0.6711409395973182,0.33557046979865535,0.11185682326622093,0.0,0.0,0.0,7.829977628635347,4.138702460850116],"type":"scatter"}], {"template":{"data":{"histogram2dcontour":[{"type":"histogram2dcontour","colorbar":{"outlinewidth":0,"ticks":""},"colorscale":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]]}],"choropleth":[{"type":"choropleth","colorbar":{"outlinewidth":0,"ticks":""}}],"histogram2d":[{"type":"histogram2d","colorbar":{"outlinewidth":0,"ticks":""},"colorscale":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]]}],"heatmap":[{"type":"heatmap","colorbar":{"outlinewidth":0,"ticks":""},"colorscale":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]]}],"heatmapgl":[{"type":"heatmapgl","colorbar":{"outlinewidth":0,"ticks":""},"colorscale":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]]}],"contourcarpet":[{"type":"contourcarpet","colorbar":{"outlinewidth":0,"ticks":""}}],"contour":[{"type":"contour","colorbar":{"outlinewidth":0,"ticks":""},"colorscale":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]]}],"surface":[{"type":"surface","colorbar":{"outlinewidth":0,"ticks":""},"colorscale":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]]}],"mesh3d":[{"type":"mesh3d","colorbar":{"outlinewidth":0,"ticks":""}}],"scatter":[{"fillpattern":{"fillmode":"overlay","size":10,"solidity":0.2},"type":"scatter"}],"parcoords":[{"type":"parcoords","line":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"scatterpolargl":[{"type":"scatterpolargl","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"bar":[{"error_x":{"color":"#2a3f5f"},"error_y":{"color":"#2a3f5f"},"marker":{"line":{"color":"#E5ECF6","width":0.5},"pattern":{"fillmode":"overlay","size":10,"solidity":0.2}},"type":"bar"}],"scattergeo":[{"type":"scattergeo","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"scatterpolar":[{"type":"scatterpolar","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"histogram":[{"marker":{"pattern":{"fillmode":"overlay","size":10,"solidity":0.2}},"type":"histogram"}],"scattergl":[{"type":"scattergl","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"scatter3d":[{"type":"scatter3d","line":{"colorbar":{"outlinewidth":0,"ticks":""}},"marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"scattermapbox":[{"type":"scattermapbox","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"scatterternary":[{"type":"scatterternary","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"scattercarpet":[{"type":"scattercarpet","marker":{"colorbar":{"outlinewidth":0,"ticks":""}}}],"carpet":[{"aaxis":{"endlinecolor":"#2a3f5f","gridcolor":"white","linecolor":"white","minorgridcolor":"white","startlinecolor":"#2a3f5f"},"baxis":{"endlinecolor":"#2a3f5f","gridcolor":"white","linecolor":"white","minorgridcolor":"white","startlinecolor":"#2a3f5f"},"type":"carpet"}],"table":[{"cells":{"fill":{"color":"#EBF0F8"},"line":{"color":"white"}},"header":{"fill":{"color":"#C8D4E3"},"line":{"color":"white"}},"type":"table"}],"barpolar":[{"marker":{"line":{"color":"#E5ECF6","width":0.5},"pattern":{"fillmode":"overlay","size":10,"solidity":0.2}},"type":"barpolar"}],"pie":[{"automargin":true,"type":"pie"}]},"layout":{"autotypenumbers":"strict","colorway":["#636efa","#EF553B","#00cc96","#ab63fa","#FFA15A","#19d3f3","#FF6692","#B6E880","#FF97FF","#FECB52"],"font":{"color":"#2a3f5f"},"hovermode":"closest","hoverlabel":{"align":"left"},"paper_bgcolor":"white","plot_bgcolor":"#E5ECF6","polar":{"bgcolor":"#E5ECF6","angularaxis":{"gridcolor":"white","linecolor":"white","ticks":""},"radialaxis":{"gridcolor":"white","linecolor":"white","ticks":""}},"ternary":{"bgcolor":"#E5ECF6","aaxis":{"gridcolor":"white","linecolor":"white","ticks":""},"baxis":{"gridcolor":"white","linecolor":"white","ticks":""},"caxis":{"gridcolor":"white","linecolor":"white","ticks":""}},"coloraxis":{"colorbar":{"outlinewidth":0,"ticks":""}},"colorscale":{"sequential":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]],"sequentialminus":[[0.0,"#0d0887"],[0.1111111111111111,"#46039f"],[0.2222222222222222,"#7201a8"],[0.3333333333333333,"#9c179e"],[0.4444444444444444,"#bd3786"],[0.5555555555555556,"#d8576b"],[0.6666666666666666,"#ed7953"],[0.7777777777777778,"#fb9f3a"],[0.8888888888888888,"#fdca26"],[1.0,"#f0f921"]],"diverging":[[0,"#8e0152"],[0.1,"#c51b7d"],[0.2,"#de77ae"],[0.3,"#f1b6da"],[0.4,"#fde0ef"],[0.5,"#f7f7f7"],[0.6,"#e6f5d0"],[0.7,"#b8e186"],[0.8,"#7fbc41"],[0.9,"#4d9221"],[1,"#276419"]]},"xaxis":{"gridcolor":"white","linecolor":"white","ticks":"","title":{"standoff":15},"zerolinecolor":"white","automargin":true,"zerolinewidth":2},"yaxis":{"gridcolor":"white","linecolor":"white","ticks":"","title":{"standoff":15},"zerolinecolor":"white","automargin":true,"zerolinewidth":2},"scene":{"xaxis":{"backgroundcolor":"#E5ECF6","gridcolor":"white","linecolor":"white","showbackground":true,"ticks":"","zerolinecolor":"white","gridwidth":2},"yaxis":{"backgroundcolor":"#E5ECF6","gridcolor":"white","linecolor":"white","showbackground":true,"ticks":"","zerolinecolor":"white","gridwidth":2},"zaxis":{"backgroundcolor":"#E5ECF6","gridcolor":"white","linecolor":"white","showbackground":true,"ticks":"","zerolinecolor":"white","gridwidth":2}},"shapedefaults":{"line":{"color":"#2a3f5f"}},"annotationdefaults":{"arrowcolor":"#2a3f5f","arrowhead":0,"arrowwidth":1},"geo":{"bgcolor":"white","landcolor":"#E5ECF6","subunitcolor":"white","showland":true,"showlakes":true,"lakecolor":"white"},"title":{"x":0.05},"mapbox":{"style":"light"}}},"legend":{"x":0,"y":1,"traceorder":"normal"},"title":{"text":"New scores as functions of old evaluations"},"yaxis":{"title":{"text":"v2 scores"}},"xaxis":{"title":{"text":"v1 scores"}}}, {"responsive": true} ) }; </script> </div> |