new evals

data/code_eval_board.csv

@@ -22,4 +22,11 @@ T,Models,ARC,HellaSwag,MMLU,TruthfulQA,Winogrande,GSM8K,Reference Model
 🔶,HuggingFaceH4/zephyr-7b-beta,0.06,0.15,0.18,0.37,0.0,0.82,mistralai/Mistral-7B-v0.1
 🔶,berkeley-nest/Starling-LM-7B-alpha,0.1,0.13,0.19,0.39,0.0,0.97,mistralai/Mistral-7B-v0.1
 🔶,Sao10K/Ana-v1-m7,0.11,0.12,0.19,0.41,0.0,0.84,mistralai/Mistral-7B-v0.1
+🔶,Open-Orca/Mistral-7B-OpenOrca,0.08,0.14,0.17,0.36,0.0,0.92,mistralai/Mistral-7B-v0.1
+🔶,jondurbin/bagel-dpo-7b-v0.1,0.12,0.14,0.21,0.47,0.91,mistralai/Mistral-7B-v0.1 
+🔶,rwitz/go-bruins-v2,0.09,0.13,0.18,0.4,0.0,0.95,mistralai/Mistral-7B-v0.1 
+🔶,EmbeddedLLM/Mistral-7B-Merge-14-v0.3,0.09,0.11,0.18,0.39,0.0,0.95,mistralai/Mistral-7B-v0.1
+🔶,chargoddard/loyal-piano-m7,0.11,0.13,0.19,0.45,0.0,0.97,mistralai/Mistral-7B-v0.1
+🔶,rishiraj/CatPPT,0.09,0.12,0.19,0.44,0.0,0.98,mistralai/Mistral-7B-v0.1
+🔶,togethercomputer/RedPajama-INCITE-Instruct-3B-v1,0.08,0.12,0.19,0.43,0.0,0.77,mistralai/Mistral-7B-v0.1 
 

data/queue.csv

@@ -1,7 +1,11 @@
 Type,Model,ref_model
-🔶 finetuned,Open-Orca/Mistral-7B-OpenOrca,mistralai/Mistral-7B-v0.1
-🔶 finetuned,jondurbin/bagel-dpo-7b-v0.1,mistralai/Mistral-7B-v0.1
-🔶 finetuned,rwitz/go-bruins-v2,mistralai/Mistral-7B-v0.1
-🔶 finetuned,chargoddard/loyal-piano-m7,mistralai/Mistral-7B-v0.1
-🔶 finetuned,EmbeddedLLM/Mistral-7B-Merge-14-v0.3,mistralai/Mistral-7B-v0.1
-🔶 finetuned,rishiraj/CatPPT,mistralai/Mistral-7B-v0.1
+🔶 finetuned,lmsys/vicuna-7b-v1.5,mistralai/Mistral-7B-v0.1
+🔶 finetuned,jan-hq/trinity-v1,mistralai/Mistral-7B-v0.1
+🔶 finetuned,microsoft/Orca-2-7b,huggyllama/llama-7b
+🟢 base,huggyllama/llama-7b,mistralai/Mistral-7B-v0.1
+🟢 base,huggyllama/llama-7b,mistralai/Mistral-7B-v0.1
+🔶 finetuned,openaccess-ai-collective/DPOpenHermes-7B-v2,mistralai/Mistral-7B-v0.1
+🟢 base,tiiuae/falcon-7b-instruct,mistralai/Mistral-7B-v0.1
+🟢 base,tiiuae/falcon-7b-instruct,mistralai/Mistral-7B-v0.1
+🟢 base,01-ai/Yi-6B,mistralai/Mistral-7B-v0.1
+🔶 finetuned,NousResearch/Nous-Hermes-llama-2-7b,mistralai/Mistral-7B-v0.1