Spaces:
Runtime error
Runtime error
Clémentine
commited on
Commit
•
6c60c29
1
Parent(s):
9296ad1
add new flags
Browse files
src/leaderboard/filter_models.py
CHANGED
|
@@ -4,6 +4,7 @@ from src.display.utils import AutoEvalColumn
|
|
| 4 |
# Models which have been flagged by users as being problematic for a reason or another
|
| 5 |
# (Model name to forum discussion link)
|
| 6 |
FLAGGED_MODELS = {
|
|
|
|
| 7 |
"Voicelab/trurl-2-13b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/202",
|
| 8 |
"deepnight-research/llama-2-70B-inst": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/207",
|
| 9 |
"Aspik101/trurl-2-13b-pl-instruct_unload": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/213",
|
|
@@ -40,7 +41,54 @@ FLAGGED_MODELS = {
|
|
| 40 |
"rwitz2/pee": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
|
| 41 |
"zyh3826 / GML-Mistral-merged-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/503",
|
| 42 |
"dillfrescott/trinity-medium": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
|
| 43 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
}
|
| 45 |
|
| 46 |
# Models which have been requested by orgs to not be submitted on the leaderboard
|
|
|
|
| 4 |
# Models which have been flagged by users as being problematic for a reason or another
|
| 5 |
# (Model name to forum discussion link)
|
| 6 |
FLAGGED_MODELS = {
|
| 7 |
+
"merged": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 8 |
"Voicelab/trurl-2-13b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/202",
|
| 9 |
"deepnight-research/llama-2-70B-inst": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/207",
|
| 10 |
"Aspik101/trurl-2-13b-pl-instruct_unload": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/213",
|
|
|
|
| 41 |
"rwitz2/pee": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
|
| 42 |
"zyh3826 / GML-Mistral-merged-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/503",
|
| 43 |
"dillfrescott/trinity-medium": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
|
| 44 |
+
"udkai/Garrulus": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/526",
|
| 45 |
+
# Merges not indicated
|
| 46 |
+
"DopeorNope/SOLARC-M-10.7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 47 |
+
"DopeorNope/SOLARC-MOE-10.7Bx6": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 48 |
+
"DopeorNope/SOLARC-MOE-10.7Bx4": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 49 |
+
"gagan3012/MetaModelv2": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 50 |
+
"gagan3012/MetaModelv3": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 51 |
+
"kyujinpy/Sakura-SOLRCA-Math-Instruct-DPO-v2": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 52 |
+
"kyujinpy/Sakura-SOLAR-Instruct-DPO-v2": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 53 |
+
"kyujinpy/Sakura-SOLRCA-Math-Instruct-DPO-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 54 |
+
"kyujinpy/Sakura-SOLRCA-Instruct-DPO": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 55 |
+
"fblgit/LUNA-SOLARkrautLM-Instruct": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 56 |
+
"CultriX/MistralTrix-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 57 |
+
"samir-fama/SamirGPT-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 58 |
+
"samir-fama/FernandoGPT-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 59 |
+
"perlthoughts/Marcoroni-8x7B-v3-MoE": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 60 |
+
"rwitz/go-bruins-v2": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 61 |
+
"rwitz/go-bruins": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 62 |
+
"Walmart-the-bag/Solar-10.7B-Cato": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 63 |
+
"aqweteddy/mistral_tv-neural-marconroni": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 64 |
+
"NExtNewChattingAI/shark_tank_ai_7_b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 65 |
+
"Q-bert/MetaMath-Cybertron": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 66 |
+
"OpenPipe/mistral-ft-optimized-1227": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 67 |
+
"perlthoughts/Falkor-7b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 68 |
+
"v1olet/v1olet_merged_dpo_7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 69 |
+
"Ba2han/BruinsV2-OpHermesNeu-11B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 70 |
+
"DopeorNope/You_can_cry_Snowman-13B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 71 |
+
"PistachioAlt/Synatra-MCS-7B-v0.3-RP-Slerp": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 72 |
+
"Weyaxi/MetaMath-una-cybertron-v2-bf16-Ties": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 73 |
+
"Weyaxi/OpenHermes-2.5-neural-chat-7b-v3-2-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 74 |
+
"perlthoughts/Falkor-8x7B-MoE": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 75 |
+
"elinas/chronos007-70b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 76 |
+
"Weyaxi/MetaMath-NeuralHermes-2.5-Mistral-7B-Linear": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 77 |
+
"Weyaxi/MetaMath-neural-chat-7b-v3-2-Ties": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 78 |
+
"diffnamehard/Mistral-CatMacaroni-slerp-uncensored-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 79 |
+
"Weyaxi/neural-chat-7b-v3-1-OpenHermes-2.5-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 80 |
+
"Weyaxi/MetaMath-NeuralHermes-2.5-Mistral-7B-Ties": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 81 |
+
"Walmart-the-bag/Misted-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 82 |
+
"garage-bAInd/Camel-Platypus2-70B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 83 |
+
"Weyaxi/OpenOrca-Zephyr-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 84 |
+
"uukuguy/speechless-mistral-7b-dare-0.85": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/510",
|
| 85 |
+
"TomGrc/FusionNet_linear": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 86 |
+
"DopeorNope/SOLARC-M-10.7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 87 |
+
"cloudyu/Mixtral_11Bx2_MoE_19B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 88 |
+
"TomGrc/FusionNet": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 89 |
+
"DopeorNope/SOLARC-MOE-10.7Bx6 ": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 90 |
+
"DopeorNope/SOLARC-MOE-10.7Bx4": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 91 |
+
"gagan3012/MetaModelv2 ": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/511",
|
| 92 |
}
|
| 93 |
|
| 94 |
# Models which have been requested by orgs to not be submitted on the leaderboard
|