Upload eval_results/HuggingFaceH4/qwen-1.5-1.8b-dpo/v0.16/mmlu/results_2024-03-25T21-38-10.916331.json with huggingface_hub
Browse files
eval_results/HuggingFaceH4/qwen-1.5-1.8b-dpo/v0.16/mmlu/results_2024-03-25T21-38-10.916331.json
ADDED
|
@@ -0,0 +1,3006 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "?",
|
| 4 |
+
"num_fewshot_seeds": 1,
|
| 5 |
+
"override_batch_size": 1,
|
| 6 |
+
"max_samples": null,
|
| 7 |
+
"job_id": "",
|
| 8 |
+
"start_time": 591079.494793358,
|
| 9 |
+
"end_time": 591911.834508147,
|
| 10 |
+
"total_evaluation_time_secondes": "832.3397147889482",
|
| 11 |
+
"model_name": "HuggingFaceH4/qwen-1.5-1.8b-dpo",
|
| 12 |
+
"model_sha": "cece5dc1dbb821c6336f7d7163496c54cde603ea",
|
| 13 |
+
"model_dtype": "torch.bfloat16",
|
| 14 |
+
"model_size": "3.79 GB",
|
| 15 |
+
"config": null
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"leaderboard|mmlu:abstract_algebra|5": {
|
| 19 |
+
"acc": 0.24,
|
| 20 |
+
"acc_stderr": 0.04292346959909283
|
| 21 |
+
},
|
| 22 |
+
"leaderboard|mmlu:anatomy|5": {
|
| 23 |
+
"acc": 0.4,
|
| 24 |
+
"acc_stderr": 0.04232073695151589
|
| 25 |
+
},
|
| 26 |
+
"leaderboard|mmlu:astronomy|5": {
|
| 27 |
+
"acc": 0.5328947368421053,
|
| 28 |
+
"acc_stderr": 0.040601270352363966
|
| 29 |
+
},
|
| 30 |
+
"leaderboard|mmlu:business_ethics|5": {
|
| 31 |
+
"acc": 0.45,
|
| 32 |
+
"acc_stderr": 0.049999999999999996
|
| 33 |
+
},
|
| 34 |
+
"leaderboard|mmlu:clinical_knowledge|5": {
|
| 35 |
+
"acc": 0.5056603773584906,
|
| 36 |
+
"acc_stderr": 0.03077090076385131
|
| 37 |
+
},
|
| 38 |
+
"leaderboard|mmlu:college_biology|5": {
|
| 39 |
+
"acc": 0.4652777777777778,
|
| 40 |
+
"acc_stderr": 0.04171115858181618
|
| 41 |
+
},
|
| 42 |
+
"leaderboard|mmlu:college_chemistry|5": {
|
| 43 |
+
"acc": 0.39,
|
| 44 |
+
"acc_stderr": 0.04902071300001975
|
| 45 |
+
},
|
| 46 |
+
"leaderboard|mmlu:college_computer_science|5": {
|
| 47 |
+
"acc": 0.35,
|
| 48 |
+
"acc_stderr": 0.047937248544110196
|
| 49 |
+
},
|
| 50 |
+
"leaderboard|mmlu:college_mathematics|5": {
|
| 51 |
+
"acc": 0.27,
|
| 52 |
+
"acc_stderr": 0.0446196043338474
|
| 53 |
+
},
|
| 54 |
+
"leaderboard|mmlu:college_medicine|5": {
|
| 55 |
+
"acc": 0.45664739884393063,
|
| 56 |
+
"acc_stderr": 0.03798106566014498
|
| 57 |
+
},
|
| 58 |
+
"leaderboard|mmlu:college_physics|5": {
|
| 59 |
+
"acc": 0.29411764705882354,
|
| 60 |
+
"acc_stderr": 0.04533838195929775
|
| 61 |
+
},
|
| 62 |
+
"leaderboard|mmlu:computer_security|5": {
|
| 63 |
+
"acc": 0.58,
|
| 64 |
+
"acc_stderr": 0.049604496374885836
|
| 65 |
+
},
|
| 66 |
+
"leaderboard|mmlu:conceptual_physics|5": {
|
| 67 |
+
"acc": 0.4085106382978723,
|
| 68 |
+
"acc_stderr": 0.032134180267015755
|
| 69 |
+
},
|
| 70 |
+
"leaderboard|mmlu:econometrics|5": {
|
| 71 |
+
"acc": 0.2719298245614035,
|
| 72 |
+
"acc_stderr": 0.04185774424022056
|
| 73 |
+
},
|
| 74 |
+
"leaderboard|mmlu:electrical_engineering|5": {
|
| 75 |
+
"acc": 0.4896551724137931,
|
| 76 |
+
"acc_stderr": 0.041657747757287644
|
| 77 |
+
},
|
| 78 |
+
"leaderboard|mmlu:elementary_mathematics|5": {
|
| 79 |
+
"acc": 0.36243386243386244,
|
| 80 |
+
"acc_stderr": 0.024757473902752045
|
| 81 |
+
},
|
| 82 |
+
"leaderboard|mmlu:formal_logic|5": {
|
| 83 |
+
"acc": 0.3333333333333333,
|
| 84 |
+
"acc_stderr": 0.042163702135578345
|
| 85 |
+
},
|
| 86 |
+
"leaderboard|mmlu:global_facts|5": {
|
| 87 |
+
"acc": 0.38,
|
| 88 |
+
"acc_stderr": 0.04878317312145634
|
| 89 |
+
},
|
| 90 |
+
"leaderboard|mmlu:high_school_biology|5": {
|
| 91 |
+
"acc": 0.5451612903225806,
|
| 92 |
+
"acc_stderr": 0.028327743091561074
|
| 93 |
+
},
|
| 94 |
+
"leaderboard|mmlu:high_school_chemistry|5": {
|
| 95 |
+
"acc": 0.39901477832512317,
|
| 96 |
+
"acc_stderr": 0.03445487686264715
|
| 97 |
+
},
|
| 98 |
+
"leaderboard|mmlu:high_school_computer_science|5": {
|
| 99 |
+
"acc": 0.44,
|
| 100 |
+
"acc_stderr": 0.049888765156985884
|
| 101 |
+
},
|
| 102 |
+
"leaderboard|mmlu:high_school_european_history|5": {
|
| 103 |
+
"acc": 0.5696969696969697,
|
| 104 |
+
"acc_stderr": 0.03866225962879077
|
| 105 |
+
},
|
| 106 |
+
"leaderboard|mmlu:high_school_geography|5": {
|
| 107 |
+
"acc": 0.4797979797979798,
|
| 108 |
+
"acc_stderr": 0.03559443565563918
|
| 109 |
+
},
|
| 110 |
+
"leaderboard|mmlu:high_school_government_and_politics|5": {
|
| 111 |
+
"acc": 0.6217616580310881,
|
| 112 |
+
"acc_stderr": 0.03499807276193338
|
| 113 |
+
},
|
| 114 |
+
"leaderboard|mmlu:high_school_macroeconomics|5": {
|
| 115 |
+
"acc": 0.4256410256410256,
|
| 116 |
+
"acc_stderr": 0.025069094387296535
|
| 117 |
+
},
|
| 118 |
+
"leaderboard|mmlu:high_school_mathematics|5": {
|
| 119 |
+
"acc": 0.34444444444444444,
|
| 120 |
+
"acc_stderr": 0.028972648884844267
|
| 121 |
+
},
|
| 122 |
+
"leaderboard|mmlu:high_school_microeconomics|5": {
|
| 123 |
+
"acc": 0.4369747899159664,
|
| 124 |
+
"acc_stderr": 0.03221943636566196
|
| 125 |
+
},
|
| 126 |
+
"leaderboard|mmlu:high_school_physics|5": {
|
| 127 |
+
"acc": 0.2582781456953642,
|
| 128 |
+
"acc_stderr": 0.035737053147634576
|
| 129 |
+
},
|
| 130 |
+
"leaderboard|mmlu:high_school_psychology|5": {
|
| 131 |
+
"acc": 0.5614678899082569,
|
| 132 |
+
"acc_stderr": 0.021274713073954562
|
| 133 |
+
},
|
| 134 |
+
"leaderboard|mmlu:high_school_statistics|5": {
|
| 135 |
+
"acc": 0.3333333333333333,
|
| 136 |
+
"acc_stderr": 0.03214952147802749
|
| 137 |
+
},
|
| 138 |
+
"leaderboard|mmlu:high_school_us_history|5": {
|
| 139 |
+
"acc": 0.5245098039215687,
|
| 140 |
+
"acc_stderr": 0.03505093194348798
|
| 141 |
+
},
|
| 142 |
+
"leaderboard|mmlu:high_school_world_history|5": {
|
| 143 |
+
"acc": 0.6371308016877637,
|
| 144 |
+
"acc_stderr": 0.031299208255302136
|
| 145 |
+
},
|
| 146 |
+
"leaderboard|mmlu:human_aging|5": {
|
| 147 |
+
"acc": 0.47085201793721976,
|
| 148 |
+
"acc_stderr": 0.03350073248773404
|
| 149 |
+
},
|
| 150 |
+
"leaderboard|mmlu:human_sexuality|5": {
|
| 151 |
+
"acc": 0.5572519083969466,
|
| 152 |
+
"acc_stderr": 0.0435644720266507
|
| 153 |
+
},
|
| 154 |
+
"leaderboard|mmlu:international_law|5": {
|
| 155 |
+
"acc": 0.5950413223140496,
|
| 156 |
+
"acc_stderr": 0.04481137755942469
|
| 157 |
+
},
|
| 158 |
+
"leaderboard|mmlu:jurisprudence|5": {
|
| 159 |
+
"acc": 0.5462962962962963,
|
| 160 |
+
"acc_stderr": 0.04812917324536823
|
| 161 |
+
},
|
| 162 |
+
"leaderboard|mmlu:logical_fallacies|5": {
|
| 163 |
+
"acc": 0.4294478527607362,
|
| 164 |
+
"acc_stderr": 0.03889066619112722
|
| 165 |
+
},
|
| 166 |
+
"leaderboard|mmlu:machine_learning|5": {
|
| 167 |
+
"acc": 0.30357142857142855,
|
| 168 |
+
"acc_stderr": 0.04364226155841044
|
| 169 |
+
},
|
| 170 |
+
"leaderboard|mmlu:management|5": {
|
| 171 |
+
"acc": 0.6796116504854369,
|
| 172 |
+
"acc_stderr": 0.04620284082280041
|
| 173 |
+
},
|
| 174 |
+
"leaderboard|mmlu:marketing|5": {
|
| 175 |
+
"acc": 0.7264957264957265,
|
| 176 |
+
"acc_stderr": 0.02920254015343118
|
| 177 |
+
},
|
| 178 |
+
"leaderboard|mmlu:medical_genetics|5": {
|
| 179 |
+
"acc": 0.61,
|
| 180 |
+
"acc_stderr": 0.04902071300001974
|
| 181 |
+
},
|
| 182 |
+
"leaderboard|mmlu:miscellaneous|5": {
|
| 183 |
+
"acc": 0.5951468710089399,
|
| 184 |
+
"acc_stderr": 0.017553246467720256
|
| 185 |
+
},
|
| 186 |
+
"leaderboard|mmlu:moral_disputes|5": {
|
| 187 |
+
"acc": 0.4884393063583815,
|
| 188 |
+
"acc_stderr": 0.026911898686377903
|
| 189 |
+
},
|
| 190 |
+
"leaderboard|mmlu:moral_scenarios|5": {
|
| 191 |
+
"acc": 0.23687150837988827,
|
| 192 |
+
"acc_stderr": 0.01421957078810399
|
| 193 |
+
},
|
| 194 |
+
"leaderboard|mmlu:nutrition|5": {
|
| 195 |
+
"acc": 0.5261437908496732,
|
| 196 |
+
"acc_stderr": 0.028590752958852394
|
| 197 |
+
},
|
| 198 |
+
"leaderboard|mmlu:philosophy|5": {
|
| 199 |
+
"acc": 0.4919614147909968,
|
| 200 |
+
"acc_stderr": 0.028394421370984545
|
| 201 |
+
},
|
| 202 |
+
"leaderboard|mmlu:prehistory|5": {
|
| 203 |
+
"acc": 0.46296296296296297,
|
| 204 |
+
"acc_stderr": 0.02774431344337654
|
| 205 |
+
},
|
| 206 |
+
"leaderboard|mmlu:professional_accounting|5": {
|
| 207 |
+
"acc": 0.36524822695035464,
|
| 208 |
+
"acc_stderr": 0.028723863853281278
|
| 209 |
+
},
|
| 210 |
+
"leaderboard|mmlu:professional_law|5": {
|
| 211 |
+
"acc": 0.3494132985658409,
|
| 212 |
+
"acc_stderr": 0.01217730625278668
|
| 213 |
+
},
|
| 214 |
+
"leaderboard|mmlu:professional_medicine|5": {
|
| 215 |
+
"acc": 0.41911764705882354,
|
| 216 |
+
"acc_stderr": 0.02997280717046463
|
| 217 |
+
},
|
| 218 |
+
"leaderboard|mmlu:professional_psychology|5": {
|
| 219 |
+
"acc": 0.42320261437908496,
|
| 220 |
+
"acc_stderr": 0.01998780976948207
|
| 221 |
+
},
|
| 222 |
+
"leaderboard|mmlu:public_relations|5": {
|
| 223 |
+
"acc": 0.5363636363636364,
|
| 224 |
+
"acc_stderr": 0.04776449162396197
|
| 225 |
+
},
|
| 226 |
+
"leaderboard|mmlu:security_studies|5": {
|
| 227 |
+
"acc": 0.5346938775510204,
|
| 228 |
+
"acc_stderr": 0.03193207024425314
|
| 229 |
+
},
|
| 230 |
+
"leaderboard|mmlu:sociology|5": {
|
| 231 |
+
"acc": 0.6119402985074627,
|
| 232 |
+
"acc_stderr": 0.034457899643627506
|
| 233 |
+
},
|
| 234 |
+
"leaderboard|mmlu:us_foreign_policy|5": {
|
| 235 |
+
"acc": 0.69,
|
| 236 |
+
"acc_stderr": 0.04648231987117316
|
| 237 |
+
},
|
| 238 |
+
"leaderboard|mmlu:virology|5": {
|
| 239 |
+
"acc": 0.40963855421686746,
|
| 240 |
+
"acc_stderr": 0.03828401115079023
|
| 241 |
+
},
|
| 242 |
+
"leaderboard|mmlu:world_religions|5": {
|
| 243 |
+
"acc": 0.5614035087719298,
|
| 244 |
+
"acc_stderr": 0.038057975055904594
|
| 245 |
+
},
|
| 246 |
+
"leaderboard|mmlu:_average|5": {
|
| 247 |
+
"acc": 0.4627857789406414,
|
| 248 |
+
"acc_stderr": 0.036247392344475986
|
| 249 |
+
}
|
| 250 |
+
},
|
| 251 |
+
"versions": {
|
| 252 |
+
"leaderboard|mmlu:abstract_algebra|5": 0,
|
| 253 |
+
"leaderboard|mmlu:anatomy|5": 0,
|
| 254 |
+
"leaderboard|mmlu:astronomy|5": 0,
|
| 255 |
+
"leaderboard|mmlu:business_ethics|5": 0,
|
| 256 |
+
"leaderboard|mmlu:clinical_knowledge|5": 0,
|
| 257 |
+
"leaderboard|mmlu:college_biology|5": 0,
|
| 258 |
+
"leaderboard|mmlu:college_chemistry|5": 0,
|
| 259 |
+
"leaderboard|mmlu:college_computer_science|5": 0,
|
| 260 |
+
"leaderboard|mmlu:college_mathematics|5": 0,
|
| 261 |
+
"leaderboard|mmlu:college_medicine|5": 0,
|
| 262 |
+
"leaderboard|mmlu:college_physics|5": 0,
|
| 263 |
+
"leaderboard|mmlu:computer_security|5": 0,
|
| 264 |
+
"leaderboard|mmlu:conceptual_physics|5": 0,
|
| 265 |
+
"leaderboard|mmlu:econometrics|5": 0,
|
| 266 |
+
"leaderboard|mmlu:electrical_engineering|5": 0,
|
| 267 |
+
"leaderboard|mmlu:elementary_mathematics|5": 0,
|
| 268 |
+
"leaderboard|mmlu:formal_logic|5": 0,
|
| 269 |
+
"leaderboard|mmlu:global_facts|5": 0,
|
| 270 |
+
"leaderboard|mmlu:high_school_biology|5": 0,
|
| 271 |
+
"leaderboard|mmlu:high_school_chemistry|5": 0,
|
| 272 |
+
"leaderboard|mmlu:high_school_computer_science|5": 0,
|
| 273 |
+
"leaderboard|mmlu:high_school_european_history|5": 0,
|
| 274 |
+
"leaderboard|mmlu:high_school_geography|5": 0,
|
| 275 |
+
"leaderboard|mmlu:high_school_government_and_politics|5": 0,
|
| 276 |
+
"leaderboard|mmlu:high_school_macroeconomics|5": 0,
|
| 277 |
+
"leaderboard|mmlu:high_school_mathematics|5": 0,
|
| 278 |
+
"leaderboard|mmlu:high_school_microeconomics|5": 0,
|
| 279 |
+
"leaderboard|mmlu:high_school_physics|5": 0,
|
| 280 |
+
"leaderboard|mmlu:high_school_psychology|5": 0,
|
| 281 |
+
"leaderboard|mmlu:high_school_statistics|5": 0,
|
| 282 |
+
"leaderboard|mmlu:high_school_us_history|5": 0,
|
| 283 |
+
"leaderboard|mmlu:high_school_world_history|5": 0,
|
| 284 |
+
"leaderboard|mmlu:human_aging|5": 0,
|
| 285 |
+
"leaderboard|mmlu:human_sexuality|5": 0,
|
| 286 |
+
"leaderboard|mmlu:international_law|5": 0,
|
| 287 |
+
"leaderboard|mmlu:jurisprudence|5": 0,
|
| 288 |
+
"leaderboard|mmlu:logical_fallacies|5": 0,
|
| 289 |
+
"leaderboard|mmlu:machine_learning|5": 0,
|
| 290 |
+
"leaderboard|mmlu:management|5": 0,
|
| 291 |
+
"leaderboard|mmlu:marketing|5": 0,
|
| 292 |
+
"leaderboard|mmlu:medical_genetics|5": 0,
|
| 293 |
+
"leaderboard|mmlu:miscellaneous|5": 0,
|
| 294 |
+
"leaderboard|mmlu:moral_disputes|5": 0,
|
| 295 |
+
"leaderboard|mmlu:moral_scenarios|5": 0,
|
| 296 |
+
"leaderboard|mmlu:nutrition|5": 0,
|
| 297 |
+
"leaderboard|mmlu:philosophy|5": 0,
|
| 298 |
+
"leaderboard|mmlu:prehistory|5": 0,
|
| 299 |
+
"leaderboard|mmlu:professional_accounting|5": 0,
|
| 300 |
+
"leaderboard|mmlu:professional_law|5": 0,
|
| 301 |
+
"leaderboard|mmlu:professional_medicine|5": 0,
|
| 302 |
+
"leaderboard|mmlu:professional_psychology|5": 0,
|
| 303 |
+
"leaderboard|mmlu:public_relations|5": 0,
|
| 304 |
+
"leaderboard|mmlu:security_studies|5": 0,
|
| 305 |
+
"leaderboard|mmlu:sociology|5": 0,
|
| 306 |
+
"leaderboard|mmlu:us_foreign_policy|5": 0,
|
| 307 |
+
"leaderboard|mmlu:virology|5": 0,
|
| 308 |
+
"leaderboard|mmlu:world_religions|5": 0
|
| 309 |
+
},
|
| 310 |
+
"config_tasks": {
|
| 311 |
+
"leaderboard|mmlu:abstract_algebra": {
|
| 312 |
+
"name": "mmlu:abstract_algebra",
|
| 313 |
+
"prompt_function": "mmlu_harness",
|
| 314 |
+
"hf_repo": "lighteval/mmlu",
|
| 315 |
+
"hf_subset": "abstract_algebra",
|
| 316 |
+
"metric": [
|
| 317 |
+
"loglikelihood_acc"
|
| 318 |
+
],
|
| 319 |
+
"hf_avail_splits": [
|
| 320 |
+
"auxiliary_train",
|
| 321 |
+
"test",
|
| 322 |
+
"validation",
|
| 323 |
+
"dev"
|
| 324 |
+
],
|
| 325 |
+
"evaluation_splits": [
|
| 326 |
+
"test"
|
| 327 |
+
],
|
| 328 |
+
"few_shots_split": "dev",
|
| 329 |
+
"few_shots_select": "sequential",
|
| 330 |
+
"generation_size": 1,
|
| 331 |
+
"stop_sequence": [
|
| 332 |
+
"\n"
|
| 333 |
+
],
|
| 334 |
+
"output_regex": null,
|
| 335 |
+
"frozen": false,
|
| 336 |
+
"suite": [
|
| 337 |
+
"leaderboard",
|
| 338 |
+
"mmlu"
|
| 339 |
+
],
|
| 340 |
+
"original_num_docs": 100,
|
| 341 |
+
"effective_num_docs": 100,
|
| 342 |
+
"trust_dataset": true
|
| 343 |
+
},
|
| 344 |
+
"leaderboard|mmlu:anatomy": {
|
| 345 |
+
"name": "mmlu:anatomy",
|
| 346 |
+
"prompt_function": "mmlu_harness",
|
| 347 |
+
"hf_repo": "lighteval/mmlu",
|
| 348 |
+
"hf_subset": "anatomy",
|
| 349 |
+
"metric": [
|
| 350 |
+
"loglikelihood_acc"
|
| 351 |
+
],
|
| 352 |
+
"hf_avail_splits": [
|
| 353 |
+
"auxiliary_train",
|
| 354 |
+
"test",
|
| 355 |
+
"validation",
|
| 356 |
+
"dev"
|
| 357 |
+
],
|
| 358 |
+
"evaluation_splits": [
|
| 359 |
+
"test"
|
| 360 |
+
],
|
| 361 |
+
"few_shots_split": "dev",
|
| 362 |
+
"few_shots_select": "sequential",
|
| 363 |
+
"generation_size": 1,
|
| 364 |
+
"stop_sequence": [
|
| 365 |
+
"\n"
|
| 366 |
+
],
|
| 367 |
+
"output_regex": null,
|
| 368 |
+
"frozen": false,
|
| 369 |
+
"suite": [
|
| 370 |
+
"leaderboard",
|
| 371 |
+
"mmlu"
|
| 372 |
+
],
|
| 373 |
+
"original_num_docs": 135,
|
| 374 |
+
"effective_num_docs": 135,
|
| 375 |
+
"trust_dataset": true
|
| 376 |
+
},
|
| 377 |
+
"leaderboard|mmlu:astronomy": {
|
| 378 |
+
"name": "mmlu:astronomy",
|
| 379 |
+
"prompt_function": "mmlu_harness",
|
| 380 |
+
"hf_repo": "lighteval/mmlu",
|
| 381 |
+
"hf_subset": "astronomy",
|
| 382 |
+
"metric": [
|
| 383 |
+
"loglikelihood_acc"
|
| 384 |
+
],
|
| 385 |
+
"hf_avail_splits": [
|
| 386 |
+
"auxiliary_train",
|
| 387 |
+
"test",
|
| 388 |
+
"validation",
|
| 389 |
+
"dev"
|
| 390 |
+
],
|
| 391 |
+
"evaluation_splits": [
|
| 392 |
+
"test"
|
| 393 |
+
],
|
| 394 |
+
"few_shots_split": "dev",
|
| 395 |
+
"few_shots_select": "sequential",
|
| 396 |
+
"generation_size": 1,
|
| 397 |
+
"stop_sequence": [
|
| 398 |
+
"\n"
|
| 399 |
+
],
|
| 400 |
+
"output_regex": null,
|
| 401 |
+
"frozen": false,
|
| 402 |
+
"suite": [
|
| 403 |
+
"leaderboard",
|
| 404 |
+
"mmlu"
|
| 405 |
+
],
|
| 406 |
+
"original_num_docs": 152,
|
| 407 |
+
"effective_num_docs": 152,
|
| 408 |
+
"trust_dataset": true
|
| 409 |
+
},
|
| 410 |
+
"leaderboard|mmlu:business_ethics": {
|
| 411 |
+
"name": "mmlu:business_ethics",
|
| 412 |
+
"prompt_function": "mmlu_harness",
|
| 413 |
+
"hf_repo": "lighteval/mmlu",
|
| 414 |
+
"hf_subset": "business_ethics",
|
| 415 |
+
"metric": [
|
| 416 |
+
"loglikelihood_acc"
|
| 417 |
+
],
|
| 418 |
+
"hf_avail_splits": [
|
| 419 |
+
"auxiliary_train",
|
| 420 |
+
"test",
|
| 421 |
+
"validation",
|
| 422 |
+
"dev"
|
| 423 |
+
],
|
| 424 |
+
"evaluation_splits": [
|
| 425 |
+
"test"
|
| 426 |
+
],
|
| 427 |
+
"few_shots_split": "dev",
|
| 428 |
+
"few_shots_select": "sequential",
|
| 429 |
+
"generation_size": 1,
|
| 430 |
+
"stop_sequence": [
|
| 431 |
+
"\n"
|
| 432 |
+
],
|
| 433 |
+
"output_regex": null,
|
| 434 |
+
"frozen": false,
|
| 435 |
+
"suite": [
|
| 436 |
+
"leaderboard",
|
| 437 |
+
"mmlu"
|
| 438 |
+
],
|
| 439 |
+
"original_num_docs": 100,
|
| 440 |
+
"effective_num_docs": 100,
|
| 441 |
+
"trust_dataset": true
|
| 442 |
+
},
|
| 443 |
+
"leaderboard|mmlu:clinical_knowledge": {
|
| 444 |
+
"name": "mmlu:clinical_knowledge",
|
| 445 |
+
"prompt_function": "mmlu_harness",
|
| 446 |
+
"hf_repo": "lighteval/mmlu",
|
| 447 |
+
"hf_subset": "clinical_knowledge",
|
| 448 |
+
"metric": [
|
| 449 |
+
"loglikelihood_acc"
|
| 450 |
+
],
|
| 451 |
+
"hf_avail_splits": [
|
| 452 |
+
"auxiliary_train",
|
| 453 |
+
"test",
|
| 454 |
+
"validation",
|
| 455 |
+
"dev"
|
| 456 |
+
],
|
| 457 |
+
"evaluation_splits": [
|
| 458 |
+
"test"
|
| 459 |
+
],
|
| 460 |
+
"few_shots_split": "dev",
|
| 461 |
+
"few_shots_select": "sequential",
|
| 462 |
+
"generation_size": 1,
|
| 463 |
+
"stop_sequence": [
|
| 464 |
+
"\n"
|
| 465 |
+
],
|
| 466 |
+
"output_regex": null,
|
| 467 |
+
"frozen": false,
|
| 468 |
+
"suite": [
|
| 469 |
+
"leaderboard",
|
| 470 |
+
"mmlu"
|
| 471 |
+
],
|
| 472 |
+
"original_num_docs": 265,
|
| 473 |
+
"effective_num_docs": 265,
|
| 474 |
+
"trust_dataset": true
|
| 475 |
+
},
|
| 476 |
+
"leaderboard|mmlu:college_biology": {
|
| 477 |
+
"name": "mmlu:college_biology",
|
| 478 |
+
"prompt_function": "mmlu_harness",
|
| 479 |
+
"hf_repo": "lighteval/mmlu",
|
| 480 |
+
"hf_subset": "college_biology",
|
| 481 |
+
"metric": [
|
| 482 |
+
"loglikelihood_acc"
|
| 483 |
+
],
|
| 484 |
+
"hf_avail_splits": [
|
| 485 |
+
"auxiliary_train",
|
| 486 |
+
"test",
|
| 487 |
+
"validation",
|
| 488 |
+
"dev"
|
| 489 |
+
],
|
| 490 |
+
"evaluation_splits": [
|
| 491 |
+
"test"
|
| 492 |
+
],
|
| 493 |
+
"few_shots_split": "dev",
|
| 494 |
+
"few_shots_select": "sequential",
|
| 495 |
+
"generation_size": 1,
|
| 496 |
+
"stop_sequence": [
|
| 497 |
+
"\n"
|
| 498 |
+
],
|
| 499 |
+
"output_regex": null,
|
| 500 |
+
"frozen": false,
|
| 501 |
+
"suite": [
|
| 502 |
+
"leaderboard",
|
| 503 |
+
"mmlu"
|
| 504 |
+
],
|
| 505 |
+
"original_num_docs": 144,
|
| 506 |
+
"effective_num_docs": 144,
|
| 507 |
+
"trust_dataset": true
|
| 508 |
+
},
|
| 509 |
+
"leaderboard|mmlu:college_chemistry": {
|
| 510 |
+
"name": "mmlu:college_chemistry",
|
| 511 |
+
"prompt_function": "mmlu_harness",
|
| 512 |
+
"hf_repo": "lighteval/mmlu",
|
| 513 |
+
"hf_subset": "college_chemistry",
|
| 514 |
+
"metric": [
|
| 515 |
+
"loglikelihood_acc"
|
| 516 |
+
],
|
| 517 |
+
"hf_avail_splits": [
|
| 518 |
+
"auxiliary_train",
|
| 519 |
+
"test",
|
| 520 |
+
"validation",
|
| 521 |
+
"dev"
|
| 522 |
+
],
|
| 523 |
+
"evaluation_splits": [
|
| 524 |
+
"test"
|
| 525 |
+
],
|
| 526 |
+
"few_shots_split": "dev",
|
| 527 |
+
"few_shots_select": "sequential",
|
| 528 |
+
"generation_size": 1,
|
| 529 |
+
"stop_sequence": [
|
| 530 |
+
"\n"
|
| 531 |
+
],
|
| 532 |
+
"output_regex": null,
|
| 533 |
+
"frozen": false,
|
| 534 |
+
"suite": [
|
| 535 |
+
"leaderboard",
|
| 536 |
+
"mmlu"
|
| 537 |
+
],
|
| 538 |
+
"original_num_docs": 100,
|
| 539 |
+
"effective_num_docs": 100,
|
| 540 |
+
"trust_dataset": true
|
| 541 |
+
},
|
| 542 |
+
"leaderboard|mmlu:college_computer_science": {
|
| 543 |
+
"name": "mmlu:college_computer_science",
|
| 544 |
+
"prompt_function": "mmlu_harness",
|
| 545 |
+
"hf_repo": "lighteval/mmlu",
|
| 546 |
+
"hf_subset": "college_computer_science",
|
| 547 |
+
"metric": [
|
| 548 |
+
"loglikelihood_acc"
|
| 549 |
+
],
|
| 550 |
+
"hf_avail_splits": [
|
| 551 |
+
"auxiliary_train",
|
| 552 |
+
"test",
|
| 553 |
+
"validation",
|
| 554 |
+
"dev"
|
| 555 |
+
],
|
| 556 |
+
"evaluation_splits": [
|
| 557 |
+
"test"
|
| 558 |
+
],
|
| 559 |
+
"few_shots_split": "dev",
|
| 560 |
+
"few_shots_select": "sequential",
|
| 561 |
+
"generation_size": 1,
|
| 562 |
+
"stop_sequence": [
|
| 563 |
+
"\n"
|
| 564 |
+
],
|
| 565 |
+
"output_regex": null,
|
| 566 |
+
"frozen": false,
|
| 567 |
+
"suite": [
|
| 568 |
+
"leaderboard",
|
| 569 |
+
"mmlu"
|
| 570 |
+
],
|
| 571 |
+
"original_num_docs": 100,
|
| 572 |
+
"effective_num_docs": 100,
|
| 573 |
+
"trust_dataset": true
|
| 574 |
+
},
|
| 575 |
+
"leaderboard|mmlu:college_mathematics": {
|
| 576 |
+
"name": "mmlu:college_mathematics",
|
| 577 |
+
"prompt_function": "mmlu_harness",
|
| 578 |
+
"hf_repo": "lighteval/mmlu",
|
| 579 |
+
"hf_subset": "college_mathematics",
|
| 580 |
+
"metric": [
|
| 581 |
+
"loglikelihood_acc"
|
| 582 |
+
],
|
| 583 |
+
"hf_avail_splits": [
|
| 584 |
+
"auxiliary_train",
|
| 585 |
+
"test",
|
| 586 |
+
"validation",
|
| 587 |
+
"dev"
|
| 588 |
+
],
|
| 589 |
+
"evaluation_splits": [
|
| 590 |
+
"test"
|
| 591 |
+
],
|
| 592 |
+
"few_shots_split": "dev",
|
| 593 |
+
"few_shots_select": "sequential",
|
| 594 |
+
"generation_size": 1,
|
| 595 |
+
"stop_sequence": [
|
| 596 |
+
"\n"
|
| 597 |
+
],
|
| 598 |
+
"output_regex": null,
|
| 599 |
+
"frozen": false,
|
| 600 |
+
"suite": [
|
| 601 |
+
"leaderboard",
|
| 602 |
+
"mmlu"
|
| 603 |
+
],
|
| 604 |
+
"original_num_docs": 100,
|
| 605 |
+
"effective_num_docs": 100,
|
| 606 |
+
"trust_dataset": true
|
| 607 |
+
},
|
| 608 |
+
"leaderboard|mmlu:college_medicine": {
|
| 609 |
+
"name": "mmlu:college_medicine",
|
| 610 |
+
"prompt_function": "mmlu_harness",
|
| 611 |
+
"hf_repo": "lighteval/mmlu",
|
| 612 |
+
"hf_subset": "college_medicine",
|
| 613 |
+
"metric": [
|
| 614 |
+
"loglikelihood_acc"
|
| 615 |
+
],
|
| 616 |
+
"hf_avail_splits": [
|
| 617 |
+
"auxiliary_train",
|
| 618 |
+
"test",
|
| 619 |
+
"validation",
|
| 620 |
+
"dev"
|
| 621 |
+
],
|
| 622 |
+
"evaluation_splits": [
|
| 623 |
+
"test"
|
| 624 |
+
],
|
| 625 |
+
"few_shots_split": "dev",
|
| 626 |
+
"few_shots_select": "sequential",
|
| 627 |
+
"generation_size": 1,
|
| 628 |
+
"stop_sequence": [
|
| 629 |
+
"\n"
|
| 630 |
+
],
|
| 631 |
+
"output_regex": null,
|
| 632 |
+
"frozen": false,
|
| 633 |
+
"suite": [
|
| 634 |
+
"leaderboard",
|
| 635 |
+
"mmlu"
|
| 636 |
+
],
|
| 637 |
+
"original_num_docs": 173,
|
| 638 |
+
"effective_num_docs": 173,
|
| 639 |
+
"trust_dataset": true
|
| 640 |
+
},
|
| 641 |
+
"leaderboard|mmlu:college_physics": {
|
| 642 |
+
"name": "mmlu:college_physics",
|
| 643 |
+
"prompt_function": "mmlu_harness",
|
| 644 |
+
"hf_repo": "lighteval/mmlu",
|
| 645 |
+
"hf_subset": "college_physics",
|
| 646 |
+
"metric": [
|
| 647 |
+
"loglikelihood_acc"
|
| 648 |
+
],
|
| 649 |
+
"hf_avail_splits": [
|
| 650 |
+
"auxiliary_train",
|
| 651 |
+
"test",
|
| 652 |
+
"validation",
|
| 653 |
+
"dev"
|
| 654 |
+
],
|
| 655 |
+
"evaluation_splits": [
|
| 656 |
+
"test"
|
| 657 |
+
],
|
| 658 |
+
"few_shots_split": "dev",
|
| 659 |
+
"few_shots_select": "sequential",
|
| 660 |
+
"generation_size": 1,
|
| 661 |
+
"stop_sequence": [
|
| 662 |
+
"\n"
|
| 663 |
+
],
|
| 664 |
+
"output_regex": null,
|
| 665 |
+
"frozen": false,
|
| 666 |
+
"suite": [
|
| 667 |
+
"leaderboard",
|
| 668 |
+
"mmlu"
|
| 669 |
+
],
|
| 670 |
+
"original_num_docs": 102,
|
| 671 |
+
"effective_num_docs": 102,
|
| 672 |
+
"trust_dataset": true
|
| 673 |
+
},
|
| 674 |
+
"leaderboard|mmlu:computer_security": {
|
| 675 |
+
"name": "mmlu:computer_security",
|
| 676 |
+
"prompt_function": "mmlu_harness",
|
| 677 |
+
"hf_repo": "lighteval/mmlu",
|
| 678 |
+
"hf_subset": "computer_security",
|
| 679 |
+
"metric": [
|
| 680 |
+
"loglikelihood_acc"
|
| 681 |
+
],
|
| 682 |
+
"hf_avail_splits": [
|
| 683 |
+
"auxiliary_train",
|
| 684 |
+
"test",
|
| 685 |
+
"validation",
|
| 686 |
+
"dev"
|
| 687 |
+
],
|
| 688 |
+
"evaluation_splits": [
|
| 689 |
+
"test"
|
| 690 |
+
],
|
| 691 |
+
"few_shots_split": "dev",
|
| 692 |
+
"few_shots_select": "sequential",
|
| 693 |
+
"generation_size": 1,
|
| 694 |
+
"stop_sequence": [
|
| 695 |
+
"\n"
|
| 696 |
+
],
|
| 697 |
+
"output_regex": null,
|
| 698 |
+
"frozen": false,
|
| 699 |
+
"suite": [
|
| 700 |
+
"leaderboard",
|
| 701 |
+
"mmlu"
|
| 702 |
+
],
|
| 703 |
+
"original_num_docs": 100,
|
| 704 |
+
"effective_num_docs": 100,
|
| 705 |
+
"trust_dataset": true
|
| 706 |
+
},
|
| 707 |
+
"leaderboard|mmlu:conceptual_physics": {
|
| 708 |
+
"name": "mmlu:conceptual_physics",
|
| 709 |
+
"prompt_function": "mmlu_harness",
|
| 710 |
+
"hf_repo": "lighteval/mmlu",
|
| 711 |
+
"hf_subset": "conceptual_physics",
|
| 712 |
+
"metric": [
|
| 713 |
+
"loglikelihood_acc"
|
| 714 |
+
],
|
| 715 |
+
"hf_avail_splits": [
|
| 716 |
+
"auxiliary_train",
|
| 717 |
+
"test",
|
| 718 |
+
"validation",
|
| 719 |
+
"dev"
|
| 720 |
+
],
|
| 721 |
+
"evaluation_splits": [
|
| 722 |
+
"test"
|
| 723 |
+
],
|
| 724 |
+
"few_shots_split": "dev",
|
| 725 |
+
"few_shots_select": "sequential",
|
| 726 |
+
"generation_size": 1,
|
| 727 |
+
"stop_sequence": [
|
| 728 |
+
"\n"
|
| 729 |
+
],
|
| 730 |
+
"output_regex": null,
|
| 731 |
+
"frozen": false,
|
| 732 |
+
"suite": [
|
| 733 |
+
"leaderboard",
|
| 734 |
+
"mmlu"
|
| 735 |
+
],
|
| 736 |
+
"original_num_docs": 235,
|
| 737 |
+
"effective_num_docs": 235,
|
| 738 |
+
"trust_dataset": true
|
| 739 |
+
},
|
| 740 |
+
"leaderboard|mmlu:econometrics": {
|
| 741 |
+
"name": "mmlu:econometrics",
|
| 742 |
+
"prompt_function": "mmlu_harness",
|
| 743 |
+
"hf_repo": "lighteval/mmlu",
|
| 744 |
+
"hf_subset": "econometrics",
|
| 745 |
+
"metric": [
|
| 746 |
+
"loglikelihood_acc"
|
| 747 |
+
],
|
| 748 |
+
"hf_avail_splits": [
|
| 749 |
+
"auxiliary_train",
|
| 750 |
+
"test",
|
| 751 |
+
"validation",
|
| 752 |
+
"dev"
|
| 753 |
+
],
|
| 754 |
+
"evaluation_splits": [
|
| 755 |
+
"test"
|
| 756 |
+
],
|
| 757 |
+
"few_shots_split": "dev",
|
| 758 |
+
"few_shots_select": "sequential",
|
| 759 |
+
"generation_size": 1,
|
| 760 |
+
"stop_sequence": [
|
| 761 |
+
"\n"
|
| 762 |
+
],
|
| 763 |
+
"output_regex": null,
|
| 764 |
+
"frozen": false,
|
| 765 |
+
"suite": [
|
| 766 |
+
"leaderboard",
|
| 767 |
+
"mmlu"
|
| 768 |
+
],
|
| 769 |
+
"original_num_docs": 114,
|
| 770 |
+
"effective_num_docs": 114,
|
| 771 |
+
"trust_dataset": true
|
| 772 |
+
},
|
| 773 |
+
"leaderboard|mmlu:electrical_engineering": {
|
| 774 |
+
"name": "mmlu:electrical_engineering",
|
| 775 |
+
"prompt_function": "mmlu_harness",
|
| 776 |
+
"hf_repo": "lighteval/mmlu",
|
| 777 |
+
"hf_subset": "electrical_engineering",
|
| 778 |
+
"metric": [
|
| 779 |
+
"loglikelihood_acc"
|
| 780 |
+
],
|
| 781 |
+
"hf_avail_splits": [
|
| 782 |
+
"auxiliary_train",
|
| 783 |
+
"test",
|
| 784 |
+
"validation",
|
| 785 |
+
"dev"
|
| 786 |
+
],
|
| 787 |
+
"evaluation_splits": [
|
| 788 |
+
"test"
|
| 789 |
+
],
|
| 790 |
+
"few_shots_split": "dev",
|
| 791 |
+
"few_shots_select": "sequential",
|
| 792 |
+
"generation_size": 1,
|
| 793 |
+
"stop_sequence": [
|
| 794 |
+
"\n"
|
| 795 |
+
],
|
| 796 |
+
"output_regex": null,
|
| 797 |
+
"frozen": false,
|
| 798 |
+
"suite": [
|
| 799 |
+
"leaderboard",
|
| 800 |
+
"mmlu"
|
| 801 |
+
],
|
| 802 |
+
"original_num_docs": 145,
|
| 803 |
+
"effective_num_docs": 145,
|
| 804 |
+
"trust_dataset": true
|
| 805 |
+
},
|
| 806 |
+
"leaderboard|mmlu:elementary_mathematics": {
|
| 807 |
+
"name": "mmlu:elementary_mathematics",
|
| 808 |
+
"prompt_function": "mmlu_harness",
|
| 809 |
+
"hf_repo": "lighteval/mmlu",
|
| 810 |
+
"hf_subset": "elementary_mathematics",
|
| 811 |
+
"metric": [
|
| 812 |
+
"loglikelihood_acc"
|
| 813 |
+
],
|
| 814 |
+
"hf_avail_splits": [
|
| 815 |
+
"auxiliary_train",
|
| 816 |
+
"test",
|
| 817 |
+
"validation",
|
| 818 |
+
"dev"
|
| 819 |
+
],
|
| 820 |
+
"evaluation_splits": [
|
| 821 |
+
"test"
|
| 822 |
+
],
|
| 823 |
+
"few_shots_split": "dev",
|
| 824 |
+
"few_shots_select": "sequential",
|
| 825 |
+
"generation_size": 1,
|
| 826 |
+
"stop_sequence": [
|
| 827 |
+
"\n"
|
| 828 |
+
],
|
| 829 |
+
"output_regex": null,
|
| 830 |
+
"frozen": false,
|
| 831 |
+
"suite": [
|
| 832 |
+
"leaderboard",
|
| 833 |
+
"mmlu"
|
| 834 |
+
],
|
| 835 |
+
"original_num_docs": 378,
|
| 836 |
+
"effective_num_docs": 378,
|
| 837 |
+
"trust_dataset": true
|
| 838 |
+
},
|
| 839 |
+
"leaderboard|mmlu:formal_logic": {
|
| 840 |
+
"name": "mmlu:formal_logic",
|
| 841 |
+
"prompt_function": "mmlu_harness",
|
| 842 |
+
"hf_repo": "lighteval/mmlu",
|
| 843 |
+
"hf_subset": "formal_logic",
|
| 844 |
+
"metric": [
|
| 845 |
+
"loglikelihood_acc"
|
| 846 |
+
],
|
| 847 |
+
"hf_avail_splits": [
|
| 848 |
+
"auxiliary_train",
|
| 849 |
+
"test",
|
| 850 |
+
"validation",
|
| 851 |
+
"dev"
|
| 852 |
+
],
|
| 853 |
+
"evaluation_splits": [
|
| 854 |
+
"test"
|
| 855 |
+
],
|
| 856 |
+
"few_shots_split": "dev",
|
| 857 |
+
"few_shots_select": "sequential",
|
| 858 |
+
"generation_size": 1,
|
| 859 |
+
"stop_sequence": [
|
| 860 |
+
"\n"
|
| 861 |
+
],
|
| 862 |
+
"output_regex": null,
|
| 863 |
+
"frozen": false,
|
| 864 |
+
"suite": [
|
| 865 |
+
"leaderboard",
|
| 866 |
+
"mmlu"
|
| 867 |
+
],
|
| 868 |
+
"original_num_docs": 126,
|
| 869 |
+
"effective_num_docs": 126,
|
| 870 |
+
"trust_dataset": true
|
| 871 |
+
},
|
| 872 |
+
"leaderboard|mmlu:global_facts": {
|
| 873 |
+
"name": "mmlu:global_facts",
|
| 874 |
+
"prompt_function": "mmlu_harness",
|
| 875 |
+
"hf_repo": "lighteval/mmlu",
|
| 876 |
+
"hf_subset": "global_facts",
|
| 877 |
+
"metric": [
|
| 878 |
+
"loglikelihood_acc"
|
| 879 |
+
],
|
| 880 |
+
"hf_avail_splits": [
|
| 881 |
+
"auxiliary_train",
|
| 882 |
+
"test",
|
| 883 |
+
"validation",
|
| 884 |
+
"dev"
|
| 885 |
+
],
|
| 886 |
+
"evaluation_splits": [
|
| 887 |
+
"test"
|
| 888 |
+
],
|
| 889 |
+
"few_shots_split": "dev",
|
| 890 |
+
"few_shots_select": "sequential",
|
| 891 |
+
"generation_size": 1,
|
| 892 |
+
"stop_sequence": [
|
| 893 |
+
"\n"
|
| 894 |
+
],
|
| 895 |
+
"output_regex": null,
|
| 896 |
+
"frozen": false,
|
| 897 |
+
"suite": [
|
| 898 |
+
"leaderboard",
|
| 899 |
+
"mmlu"
|
| 900 |
+
],
|
| 901 |
+
"original_num_docs": 100,
|
| 902 |
+
"effective_num_docs": 100,
|
| 903 |
+
"trust_dataset": true
|
| 904 |
+
},
|
| 905 |
+
"leaderboard|mmlu:high_school_biology": {
|
| 906 |
+
"name": "mmlu:high_school_biology",
|
| 907 |
+
"prompt_function": "mmlu_harness",
|
| 908 |
+
"hf_repo": "lighteval/mmlu",
|
| 909 |
+
"hf_subset": "high_school_biology",
|
| 910 |
+
"metric": [
|
| 911 |
+
"loglikelihood_acc"
|
| 912 |
+
],
|
| 913 |
+
"hf_avail_splits": [
|
| 914 |
+
"auxiliary_train",
|
| 915 |
+
"test",
|
| 916 |
+
"validation",
|
| 917 |
+
"dev"
|
| 918 |
+
],
|
| 919 |
+
"evaluation_splits": [
|
| 920 |
+
"test"
|
| 921 |
+
],
|
| 922 |
+
"few_shots_split": "dev",
|
| 923 |
+
"few_shots_select": "sequential",
|
| 924 |
+
"generation_size": 1,
|
| 925 |
+
"stop_sequence": [
|
| 926 |
+
"\n"
|
| 927 |
+
],
|
| 928 |
+
"output_regex": null,
|
| 929 |
+
"frozen": false,
|
| 930 |
+
"suite": [
|
| 931 |
+
"leaderboard",
|
| 932 |
+
"mmlu"
|
| 933 |
+
],
|
| 934 |
+
"original_num_docs": 310,
|
| 935 |
+
"effective_num_docs": 310,
|
| 936 |
+
"trust_dataset": true
|
| 937 |
+
},
|
| 938 |
+
"leaderboard|mmlu:high_school_chemistry": {
|
| 939 |
+
"name": "mmlu:high_school_chemistry",
|
| 940 |
+
"prompt_function": "mmlu_harness",
|
| 941 |
+
"hf_repo": "lighteval/mmlu",
|
| 942 |
+
"hf_subset": "high_school_chemistry",
|
| 943 |
+
"metric": [
|
| 944 |
+
"loglikelihood_acc"
|
| 945 |
+
],
|
| 946 |
+
"hf_avail_splits": [
|
| 947 |
+
"auxiliary_train",
|
| 948 |
+
"test",
|
| 949 |
+
"validation",
|
| 950 |
+
"dev"
|
| 951 |
+
],
|
| 952 |
+
"evaluation_splits": [
|
| 953 |
+
"test"
|
| 954 |
+
],
|
| 955 |
+
"few_shots_split": "dev",
|
| 956 |
+
"few_shots_select": "sequential",
|
| 957 |
+
"generation_size": 1,
|
| 958 |
+
"stop_sequence": [
|
| 959 |
+
"\n"
|
| 960 |
+
],
|
| 961 |
+
"output_regex": null,
|
| 962 |
+
"frozen": false,
|
| 963 |
+
"suite": [
|
| 964 |
+
"leaderboard",
|
| 965 |
+
"mmlu"
|
| 966 |
+
],
|
| 967 |
+
"original_num_docs": 203,
|
| 968 |
+
"effective_num_docs": 203,
|
| 969 |
+
"trust_dataset": true
|
| 970 |
+
},
|
| 971 |
+
"leaderboard|mmlu:high_school_computer_science": {
|
| 972 |
+
"name": "mmlu:high_school_computer_science",
|
| 973 |
+
"prompt_function": "mmlu_harness",
|
| 974 |
+
"hf_repo": "lighteval/mmlu",
|
| 975 |
+
"hf_subset": "high_school_computer_science",
|
| 976 |
+
"metric": [
|
| 977 |
+
"loglikelihood_acc"
|
| 978 |
+
],
|
| 979 |
+
"hf_avail_splits": [
|
| 980 |
+
"auxiliary_train",
|
| 981 |
+
"test",
|
| 982 |
+
"validation",
|
| 983 |
+
"dev"
|
| 984 |
+
],
|
| 985 |
+
"evaluation_splits": [
|
| 986 |
+
"test"
|
| 987 |
+
],
|
| 988 |
+
"few_shots_split": "dev",
|
| 989 |
+
"few_shots_select": "sequential",
|
| 990 |
+
"generation_size": 1,
|
| 991 |
+
"stop_sequence": [
|
| 992 |
+
"\n"
|
| 993 |
+
],
|
| 994 |
+
"output_regex": null,
|
| 995 |
+
"frozen": false,
|
| 996 |
+
"suite": [
|
| 997 |
+
"leaderboard",
|
| 998 |
+
"mmlu"
|
| 999 |
+
],
|
| 1000 |
+
"original_num_docs": 100,
|
| 1001 |
+
"effective_num_docs": 100,
|
| 1002 |
+
"trust_dataset": true
|
| 1003 |
+
},
|
| 1004 |
+
"leaderboard|mmlu:high_school_european_history": {
|
| 1005 |
+
"name": "mmlu:high_school_european_history",
|
| 1006 |
+
"prompt_function": "mmlu_harness",
|
| 1007 |
+
"hf_repo": "lighteval/mmlu",
|
| 1008 |
+
"hf_subset": "high_school_european_history",
|
| 1009 |
+
"metric": [
|
| 1010 |
+
"loglikelihood_acc"
|
| 1011 |
+
],
|
| 1012 |
+
"hf_avail_splits": [
|
| 1013 |
+
"auxiliary_train",
|
| 1014 |
+
"test",
|
| 1015 |
+
"validation",
|
| 1016 |
+
"dev"
|
| 1017 |
+
],
|
| 1018 |
+
"evaluation_splits": [
|
| 1019 |
+
"test"
|
| 1020 |
+
],
|
| 1021 |
+
"few_shots_split": "dev",
|
| 1022 |
+
"few_shots_select": "sequential",
|
| 1023 |
+
"generation_size": 1,
|
| 1024 |
+
"stop_sequence": [
|
| 1025 |
+
"\n"
|
| 1026 |
+
],
|
| 1027 |
+
"output_regex": null,
|
| 1028 |
+
"frozen": false,
|
| 1029 |
+
"suite": [
|
| 1030 |
+
"leaderboard",
|
| 1031 |
+
"mmlu"
|
| 1032 |
+
],
|
| 1033 |
+
"original_num_docs": 165,
|
| 1034 |
+
"effective_num_docs": 165,
|
| 1035 |
+
"trust_dataset": true
|
| 1036 |
+
},
|
| 1037 |
+
"leaderboard|mmlu:high_school_geography": {
|
| 1038 |
+
"name": "mmlu:high_school_geography",
|
| 1039 |
+
"prompt_function": "mmlu_harness",
|
| 1040 |
+
"hf_repo": "lighteval/mmlu",
|
| 1041 |
+
"hf_subset": "high_school_geography",
|
| 1042 |
+
"metric": [
|
| 1043 |
+
"loglikelihood_acc"
|
| 1044 |
+
],
|
| 1045 |
+
"hf_avail_splits": [
|
| 1046 |
+
"auxiliary_train",
|
| 1047 |
+
"test",
|
| 1048 |
+
"validation",
|
| 1049 |
+
"dev"
|
| 1050 |
+
],
|
| 1051 |
+
"evaluation_splits": [
|
| 1052 |
+
"test"
|
| 1053 |
+
],
|
| 1054 |
+
"few_shots_split": "dev",
|
| 1055 |
+
"few_shots_select": "sequential",
|
| 1056 |
+
"generation_size": 1,
|
| 1057 |
+
"stop_sequence": [
|
| 1058 |
+
"\n"
|
| 1059 |
+
],
|
| 1060 |
+
"output_regex": null,
|
| 1061 |
+
"frozen": false,
|
| 1062 |
+
"suite": [
|
| 1063 |
+
"leaderboard",
|
| 1064 |
+
"mmlu"
|
| 1065 |
+
],
|
| 1066 |
+
"original_num_docs": 198,
|
| 1067 |
+
"effective_num_docs": 198,
|
| 1068 |
+
"trust_dataset": true
|
| 1069 |
+
},
|
| 1070 |
+
"leaderboard|mmlu:high_school_government_and_politics": {
|
| 1071 |
+
"name": "mmlu:high_school_government_and_politics",
|
| 1072 |
+
"prompt_function": "mmlu_harness",
|
| 1073 |
+
"hf_repo": "lighteval/mmlu",
|
| 1074 |
+
"hf_subset": "high_school_government_and_politics",
|
| 1075 |
+
"metric": [
|
| 1076 |
+
"loglikelihood_acc"
|
| 1077 |
+
],
|
| 1078 |
+
"hf_avail_splits": [
|
| 1079 |
+
"auxiliary_train",
|
| 1080 |
+
"test",
|
| 1081 |
+
"validation",
|
| 1082 |
+
"dev"
|
| 1083 |
+
],
|
| 1084 |
+
"evaluation_splits": [
|
| 1085 |
+
"test"
|
| 1086 |
+
],
|
| 1087 |
+
"few_shots_split": "dev",
|
| 1088 |
+
"few_shots_select": "sequential",
|
| 1089 |
+
"generation_size": 1,
|
| 1090 |
+
"stop_sequence": [
|
| 1091 |
+
"\n"
|
| 1092 |
+
],
|
| 1093 |
+
"output_regex": null,
|
| 1094 |
+
"frozen": false,
|
| 1095 |
+
"suite": [
|
| 1096 |
+
"leaderboard",
|
| 1097 |
+
"mmlu"
|
| 1098 |
+
],
|
| 1099 |
+
"original_num_docs": 193,
|
| 1100 |
+
"effective_num_docs": 193,
|
| 1101 |
+
"trust_dataset": true
|
| 1102 |
+
},
|
| 1103 |
+
"leaderboard|mmlu:high_school_macroeconomics": {
|
| 1104 |
+
"name": "mmlu:high_school_macroeconomics",
|
| 1105 |
+
"prompt_function": "mmlu_harness",
|
| 1106 |
+
"hf_repo": "lighteval/mmlu",
|
| 1107 |
+
"hf_subset": "high_school_macroeconomics",
|
| 1108 |
+
"metric": [
|
| 1109 |
+
"loglikelihood_acc"
|
| 1110 |
+
],
|
| 1111 |
+
"hf_avail_splits": [
|
| 1112 |
+
"auxiliary_train",
|
| 1113 |
+
"test",
|
| 1114 |
+
"validation",
|
| 1115 |
+
"dev"
|
| 1116 |
+
],
|
| 1117 |
+
"evaluation_splits": [
|
| 1118 |
+
"test"
|
| 1119 |
+
],
|
| 1120 |
+
"few_shots_split": "dev",
|
| 1121 |
+
"few_shots_select": "sequential",
|
| 1122 |
+
"generation_size": 1,
|
| 1123 |
+
"stop_sequence": [
|
| 1124 |
+
"\n"
|
| 1125 |
+
],
|
| 1126 |
+
"output_regex": null,
|
| 1127 |
+
"frozen": false,
|
| 1128 |
+
"suite": [
|
| 1129 |
+
"leaderboard",
|
| 1130 |
+
"mmlu"
|
| 1131 |
+
],
|
| 1132 |
+
"original_num_docs": 390,
|
| 1133 |
+
"effective_num_docs": 390,
|
| 1134 |
+
"trust_dataset": true
|
| 1135 |
+
},
|
| 1136 |
+
"leaderboard|mmlu:high_school_mathematics": {
|
| 1137 |
+
"name": "mmlu:high_school_mathematics",
|
| 1138 |
+
"prompt_function": "mmlu_harness",
|
| 1139 |
+
"hf_repo": "lighteval/mmlu",
|
| 1140 |
+
"hf_subset": "high_school_mathematics",
|
| 1141 |
+
"metric": [
|
| 1142 |
+
"loglikelihood_acc"
|
| 1143 |
+
],
|
| 1144 |
+
"hf_avail_splits": [
|
| 1145 |
+
"auxiliary_train",
|
| 1146 |
+
"test",
|
| 1147 |
+
"validation",
|
| 1148 |
+
"dev"
|
| 1149 |
+
],
|
| 1150 |
+
"evaluation_splits": [
|
| 1151 |
+
"test"
|
| 1152 |
+
],
|
| 1153 |
+
"few_shots_split": "dev",
|
| 1154 |
+
"few_shots_select": "sequential",
|
| 1155 |
+
"generation_size": 1,
|
| 1156 |
+
"stop_sequence": [
|
| 1157 |
+
"\n"
|
| 1158 |
+
],
|
| 1159 |
+
"output_regex": null,
|
| 1160 |
+
"frozen": false,
|
| 1161 |
+
"suite": [
|
| 1162 |
+
"leaderboard",
|
| 1163 |
+
"mmlu"
|
| 1164 |
+
],
|
| 1165 |
+
"original_num_docs": 270,
|
| 1166 |
+
"effective_num_docs": 270,
|
| 1167 |
+
"trust_dataset": true
|
| 1168 |
+
},
|
| 1169 |
+
"leaderboard|mmlu:high_school_microeconomics": {
|
| 1170 |
+
"name": "mmlu:high_school_microeconomics",
|
| 1171 |
+
"prompt_function": "mmlu_harness",
|
| 1172 |
+
"hf_repo": "lighteval/mmlu",
|
| 1173 |
+
"hf_subset": "high_school_microeconomics",
|
| 1174 |
+
"metric": [
|
| 1175 |
+
"loglikelihood_acc"
|
| 1176 |
+
],
|
| 1177 |
+
"hf_avail_splits": [
|
| 1178 |
+
"auxiliary_train",
|
| 1179 |
+
"test",
|
| 1180 |
+
"validation",
|
| 1181 |
+
"dev"
|
| 1182 |
+
],
|
| 1183 |
+
"evaluation_splits": [
|
| 1184 |
+
"test"
|
| 1185 |
+
],
|
| 1186 |
+
"few_shots_split": "dev",
|
| 1187 |
+
"few_shots_select": "sequential",
|
| 1188 |
+
"generation_size": 1,
|
| 1189 |
+
"stop_sequence": [
|
| 1190 |
+
"\n"
|
| 1191 |
+
],
|
| 1192 |
+
"output_regex": null,
|
| 1193 |
+
"frozen": false,
|
| 1194 |
+
"suite": [
|
| 1195 |
+
"leaderboard",
|
| 1196 |
+
"mmlu"
|
| 1197 |
+
],
|
| 1198 |
+
"original_num_docs": 238,
|
| 1199 |
+
"effective_num_docs": 238,
|
| 1200 |
+
"trust_dataset": true
|
| 1201 |
+
},
|
| 1202 |
+
"leaderboard|mmlu:high_school_physics": {
|
| 1203 |
+
"name": "mmlu:high_school_physics",
|
| 1204 |
+
"prompt_function": "mmlu_harness",
|
| 1205 |
+
"hf_repo": "lighteval/mmlu",
|
| 1206 |
+
"hf_subset": "high_school_physics",
|
| 1207 |
+
"metric": [
|
| 1208 |
+
"loglikelihood_acc"
|
| 1209 |
+
],
|
| 1210 |
+
"hf_avail_splits": [
|
| 1211 |
+
"auxiliary_train",
|
| 1212 |
+
"test",
|
| 1213 |
+
"validation",
|
| 1214 |
+
"dev"
|
| 1215 |
+
],
|
| 1216 |
+
"evaluation_splits": [
|
| 1217 |
+
"test"
|
| 1218 |
+
],
|
| 1219 |
+
"few_shots_split": "dev",
|
| 1220 |
+
"few_shots_select": "sequential",
|
| 1221 |
+
"generation_size": 1,
|
| 1222 |
+
"stop_sequence": [
|
| 1223 |
+
"\n"
|
| 1224 |
+
],
|
| 1225 |
+
"output_regex": null,
|
| 1226 |
+
"frozen": false,
|
| 1227 |
+
"suite": [
|
| 1228 |
+
"leaderboard",
|
| 1229 |
+
"mmlu"
|
| 1230 |
+
],
|
| 1231 |
+
"original_num_docs": 151,
|
| 1232 |
+
"effective_num_docs": 151,
|
| 1233 |
+
"trust_dataset": true
|
| 1234 |
+
},
|
| 1235 |
+
"leaderboard|mmlu:high_school_psychology": {
|
| 1236 |
+
"name": "mmlu:high_school_psychology",
|
| 1237 |
+
"prompt_function": "mmlu_harness",
|
| 1238 |
+
"hf_repo": "lighteval/mmlu",
|
| 1239 |
+
"hf_subset": "high_school_psychology",
|
| 1240 |
+
"metric": [
|
| 1241 |
+
"loglikelihood_acc"
|
| 1242 |
+
],
|
| 1243 |
+
"hf_avail_splits": [
|
| 1244 |
+
"auxiliary_train",
|
| 1245 |
+
"test",
|
| 1246 |
+
"validation",
|
| 1247 |
+
"dev"
|
| 1248 |
+
],
|
| 1249 |
+
"evaluation_splits": [
|
| 1250 |
+
"test"
|
| 1251 |
+
],
|
| 1252 |
+
"few_shots_split": "dev",
|
| 1253 |
+
"few_shots_select": "sequential",
|
| 1254 |
+
"generation_size": 1,
|
| 1255 |
+
"stop_sequence": [
|
| 1256 |
+
"\n"
|
| 1257 |
+
],
|
| 1258 |
+
"output_regex": null,
|
| 1259 |
+
"frozen": false,
|
| 1260 |
+
"suite": [
|
| 1261 |
+
"leaderboard",
|
| 1262 |
+
"mmlu"
|
| 1263 |
+
],
|
| 1264 |
+
"original_num_docs": 545,
|
| 1265 |
+
"effective_num_docs": 545,
|
| 1266 |
+
"trust_dataset": true
|
| 1267 |
+
},
|
| 1268 |
+
"leaderboard|mmlu:high_school_statistics": {
|
| 1269 |
+
"name": "mmlu:high_school_statistics",
|
| 1270 |
+
"prompt_function": "mmlu_harness",
|
| 1271 |
+
"hf_repo": "lighteval/mmlu",
|
| 1272 |
+
"hf_subset": "high_school_statistics",
|
| 1273 |
+
"metric": [
|
| 1274 |
+
"loglikelihood_acc"
|
| 1275 |
+
],
|
| 1276 |
+
"hf_avail_splits": [
|
| 1277 |
+
"auxiliary_train",
|
| 1278 |
+
"test",
|
| 1279 |
+
"validation",
|
| 1280 |
+
"dev"
|
| 1281 |
+
],
|
| 1282 |
+
"evaluation_splits": [
|
| 1283 |
+
"test"
|
| 1284 |
+
],
|
| 1285 |
+
"few_shots_split": "dev",
|
| 1286 |
+
"few_shots_select": "sequential",
|
| 1287 |
+
"generation_size": 1,
|
| 1288 |
+
"stop_sequence": [
|
| 1289 |
+
"\n"
|
| 1290 |
+
],
|
| 1291 |
+
"output_regex": null,
|
| 1292 |
+
"frozen": false,
|
| 1293 |
+
"suite": [
|
| 1294 |
+
"leaderboard",
|
| 1295 |
+
"mmlu"
|
| 1296 |
+
],
|
| 1297 |
+
"original_num_docs": 216,
|
| 1298 |
+
"effective_num_docs": 216,
|
| 1299 |
+
"trust_dataset": true
|
| 1300 |
+
},
|
| 1301 |
+
"leaderboard|mmlu:high_school_us_history": {
|
| 1302 |
+
"name": "mmlu:high_school_us_history",
|
| 1303 |
+
"prompt_function": "mmlu_harness",
|
| 1304 |
+
"hf_repo": "lighteval/mmlu",
|
| 1305 |
+
"hf_subset": "high_school_us_history",
|
| 1306 |
+
"metric": [
|
| 1307 |
+
"loglikelihood_acc"
|
| 1308 |
+
],
|
| 1309 |
+
"hf_avail_splits": [
|
| 1310 |
+
"auxiliary_train",
|
| 1311 |
+
"test",
|
| 1312 |
+
"validation",
|
| 1313 |
+
"dev"
|
| 1314 |
+
],
|
| 1315 |
+
"evaluation_splits": [
|
| 1316 |
+
"test"
|
| 1317 |
+
],
|
| 1318 |
+
"few_shots_split": "dev",
|
| 1319 |
+
"few_shots_select": "sequential",
|
| 1320 |
+
"generation_size": 1,
|
| 1321 |
+
"stop_sequence": [
|
| 1322 |
+
"\n"
|
| 1323 |
+
],
|
| 1324 |
+
"output_regex": null,
|
| 1325 |
+
"frozen": false,
|
| 1326 |
+
"suite": [
|
| 1327 |
+
"leaderboard",
|
| 1328 |
+
"mmlu"
|
| 1329 |
+
],
|
| 1330 |
+
"original_num_docs": 204,
|
| 1331 |
+
"effective_num_docs": 204,
|
| 1332 |
+
"trust_dataset": true
|
| 1333 |
+
},
|
| 1334 |
+
"leaderboard|mmlu:high_school_world_history": {
|
| 1335 |
+
"name": "mmlu:high_school_world_history",
|
| 1336 |
+
"prompt_function": "mmlu_harness",
|
| 1337 |
+
"hf_repo": "lighteval/mmlu",
|
| 1338 |
+
"hf_subset": "high_school_world_history",
|
| 1339 |
+
"metric": [
|
| 1340 |
+
"loglikelihood_acc"
|
| 1341 |
+
],
|
| 1342 |
+
"hf_avail_splits": [
|
| 1343 |
+
"auxiliary_train",
|
| 1344 |
+
"test",
|
| 1345 |
+
"validation",
|
| 1346 |
+
"dev"
|
| 1347 |
+
],
|
| 1348 |
+
"evaluation_splits": [
|
| 1349 |
+
"test"
|
| 1350 |
+
],
|
| 1351 |
+
"few_shots_split": "dev",
|
| 1352 |
+
"few_shots_select": "sequential",
|
| 1353 |
+
"generation_size": 1,
|
| 1354 |
+
"stop_sequence": [
|
| 1355 |
+
"\n"
|
| 1356 |
+
],
|
| 1357 |
+
"output_regex": null,
|
| 1358 |
+
"frozen": false,
|
| 1359 |
+
"suite": [
|
| 1360 |
+
"leaderboard",
|
| 1361 |
+
"mmlu"
|
| 1362 |
+
],
|
| 1363 |
+
"original_num_docs": 237,
|
| 1364 |
+
"effective_num_docs": 237,
|
| 1365 |
+
"trust_dataset": true
|
| 1366 |
+
},
|
| 1367 |
+
"leaderboard|mmlu:human_aging": {
|
| 1368 |
+
"name": "mmlu:human_aging",
|
| 1369 |
+
"prompt_function": "mmlu_harness",
|
| 1370 |
+
"hf_repo": "lighteval/mmlu",
|
| 1371 |
+
"hf_subset": "human_aging",
|
| 1372 |
+
"metric": [
|
| 1373 |
+
"loglikelihood_acc"
|
| 1374 |
+
],
|
| 1375 |
+
"hf_avail_splits": [
|
| 1376 |
+
"auxiliary_train",
|
| 1377 |
+
"test",
|
| 1378 |
+
"validation",
|
| 1379 |
+
"dev"
|
| 1380 |
+
],
|
| 1381 |
+
"evaluation_splits": [
|
| 1382 |
+
"test"
|
| 1383 |
+
],
|
| 1384 |
+
"few_shots_split": "dev",
|
| 1385 |
+
"few_shots_select": "sequential",
|
| 1386 |
+
"generation_size": 1,
|
| 1387 |
+
"stop_sequence": [
|
| 1388 |
+
"\n"
|
| 1389 |
+
],
|
| 1390 |
+
"output_regex": null,
|
| 1391 |
+
"frozen": false,
|
| 1392 |
+
"suite": [
|
| 1393 |
+
"leaderboard",
|
| 1394 |
+
"mmlu"
|
| 1395 |
+
],
|
| 1396 |
+
"original_num_docs": 223,
|
| 1397 |
+
"effective_num_docs": 223,
|
| 1398 |
+
"trust_dataset": true
|
| 1399 |
+
},
|
| 1400 |
+
"leaderboard|mmlu:human_sexuality": {
|
| 1401 |
+
"name": "mmlu:human_sexuality",
|
| 1402 |
+
"prompt_function": "mmlu_harness",
|
| 1403 |
+
"hf_repo": "lighteval/mmlu",
|
| 1404 |
+
"hf_subset": "human_sexuality",
|
| 1405 |
+
"metric": [
|
| 1406 |
+
"loglikelihood_acc"
|
| 1407 |
+
],
|
| 1408 |
+
"hf_avail_splits": [
|
| 1409 |
+
"auxiliary_train",
|
| 1410 |
+
"test",
|
| 1411 |
+
"validation",
|
| 1412 |
+
"dev"
|
| 1413 |
+
],
|
| 1414 |
+
"evaluation_splits": [
|
| 1415 |
+
"test"
|
| 1416 |
+
],
|
| 1417 |
+
"few_shots_split": "dev",
|
| 1418 |
+
"few_shots_select": "sequential",
|
| 1419 |
+
"generation_size": 1,
|
| 1420 |
+
"stop_sequence": [
|
| 1421 |
+
"\n"
|
| 1422 |
+
],
|
| 1423 |
+
"output_regex": null,
|
| 1424 |
+
"frozen": false,
|
| 1425 |
+
"suite": [
|
| 1426 |
+
"leaderboard",
|
| 1427 |
+
"mmlu"
|
| 1428 |
+
],
|
| 1429 |
+
"original_num_docs": 131,
|
| 1430 |
+
"effective_num_docs": 131,
|
| 1431 |
+
"trust_dataset": true
|
| 1432 |
+
},
|
| 1433 |
+
"leaderboard|mmlu:international_law": {
|
| 1434 |
+
"name": "mmlu:international_law",
|
| 1435 |
+
"prompt_function": "mmlu_harness",
|
| 1436 |
+
"hf_repo": "lighteval/mmlu",
|
| 1437 |
+
"hf_subset": "international_law",
|
| 1438 |
+
"metric": [
|
| 1439 |
+
"loglikelihood_acc"
|
| 1440 |
+
],
|
| 1441 |
+
"hf_avail_splits": [
|
| 1442 |
+
"auxiliary_train",
|
| 1443 |
+
"test",
|
| 1444 |
+
"validation",
|
| 1445 |
+
"dev"
|
| 1446 |
+
],
|
| 1447 |
+
"evaluation_splits": [
|
| 1448 |
+
"test"
|
| 1449 |
+
],
|
| 1450 |
+
"few_shots_split": "dev",
|
| 1451 |
+
"few_shots_select": "sequential",
|
| 1452 |
+
"generation_size": 1,
|
| 1453 |
+
"stop_sequence": [
|
| 1454 |
+
"\n"
|
| 1455 |
+
],
|
| 1456 |
+
"output_regex": null,
|
| 1457 |
+
"frozen": false,
|
| 1458 |
+
"suite": [
|
| 1459 |
+
"leaderboard",
|
| 1460 |
+
"mmlu"
|
| 1461 |
+
],
|
| 1462 |
+
"original_num_docs": 121,
|
| 1463 |
+
"effective_num_docs": 121,
|
| 1464 |
+
"trust_dataset": true
|
| 1465 |
+
},
|
| 1466 |
+
"leaderboard|mmlu:jurisprudence": {
|
| 1467 |
+
"name": "mmlu:jurisprudence",
|
| 1468 |
+
"prompt_function": "mmlu_harness",
|
| 1469 |
+
"hf_repo": "lighteval/mmlu",
|
| 1470 |
+
"hf_subset": "jurisprudence",
|
| 1471 |
+
"metric": [
|
| 1472 |
+
"loglikelihood_acc"
|
| 1473 |
+
],
|
| 1474 |
+
"hf_avail_splits": [
|
| 1475 |
+
"auxiliary_train",
|
| 1476 |
+
"test",
|
| 1477 |
+
"validation",
|
| 1478 |
+
"dev"
|
| 1479 |
+
],
|
| 1480 |
+
"evaluation_splits": [
|
| 1481 |
+
"test"
|
| 1482 |
+
],
|
| 1483 |
+
"few_shots_split": "dev",
|
| 1484 |
+
"few_shots_select": "sequential",
|
| 1485 |
+
"generation_size": 1,
|
| 1486 |
+
"stop_sequence": [
|
| 1487 |
+
"\n"
|
| 1488 |
+
],
|
| 1489 |
+
"output_regex": null,
|
| 1490 |
+
"frozen": false,
|
| 1491 |
+
"suite": [
|
| 1492 |
+
"leaderboard",
|
| 1493 |
+
"mmlu"
|
| 1494 |
+
],
|
| 1495 |
+
"original_num_docs": 108,
|
| 1496 |
+
"effective_num_docs": 108,
|
| 1497 |
+
"trust_dataset": true
|
| 1498 |
+
},
|
| 1499 |
+
"leaderboard|mmlu:logical_fallacies": {
|
| 1500 |
+
"name": "mmlu:logical_fallacies",
|
| 1501 |
+
"prompt_function": "mmlu_harness",
|
| 1502 |
+
"hf_repo": "lighteval/mmlu",
|
| 1503 |
+
"hf_subset": "logical_fallacies",
|
| 1504 |
+
"metric": [
|
| 1505 |
+
"loglikelihood_acc"
|
| 1506 |
+
],
|
| 1507 |
+
"hf_avail_splits": [
|
| 1508 |
+
"auxiliary_train",
|
| 1509 |
+
"test",
|
| 1510 |
+
"validation",
|
| 1511 |
+
"dev"
|
| 1512 |
+
],
|
| 1513 |
+
"evaluation_splits": [
|
| 1514 |
+
"test"
|
| 1515 |
+
],
|
| 1516 |
+
"few_shots_split": "dev",
|
| 1517 |
+
"few_shots_select": "sequential",
|
| 1518 |
+
"generation_size": 1,
|
| 1519 |
+
"stop_sequence": [
|
| 1520 |
+
"\n"
|
| 1521 |
+
],
|
| 1522 |
+
"output_regex": null,
|
| 1523 |
+
"frozen": false,
|
| 1524 |
+
"suite": [
|
| 1525 |
+
"leaderboard",
|
| 1526 |
+
"mmlu"
|
| 1527 |
+
],
|
| 1528 |
+
"original_num_docs": 163,
|
| 1529 |
+
"effective_num_docs": 163,
|
| 1530 |
+
"trust_dataset": true
|
| 1531 |
+
},
|
| 1532 |
+
"leaderboard|mmlu:machine_learning": {
|
| 1533 |
+
"name": "mmlu:machine_learning",
|
| 1534 |
+
"prompt_function": "mmlu_harness",
|
| 1535 |
+
"hf_repo": "lighteval/mmlu",
|
| 1536 |
+
"hf_subset": "machine_learning",
|
| 1537 |
+
"metric": [
|
| 1538 |
+
"loglikelihood_acc"
|
| 1539 |
+
],
|
| 1540 |
+
"hf_avail_splits": [
|
| 1541 |
+
"auxiliary_train",
|
| 1542 |
+
"test",
|
| 1543 |
+
"validation",
|
| 1544 |
+
"dev"
|
| 1545 |
+
],
|
| 1546 |
+
"evaluation_splits": [
|
| 1547 |
+
"test"
|
| 1548 |
+
],
|
| 1549 |
+
"few_shots_split": "dev",
|
| 1550 |
+
"few_shots_select": "sequential",
|
| 1551 |
+
"generation_size": 1,
|
| 1552 |
+
"stop_sequence": [
|
| 1553 |
+
"\n"
|
| 1554 |
+
],
|
| 1555 |
+
"output_regex": null,
|
| 1556 |
+
"frozen": false,
|
| 1557 |
+
"suite": [
|
| 1558 |
+
"leaderboard",
|
| 1559 |
+
"mmlu"
|
| 1560 |
+
],
|
| 1561 |
+
"original_num_docs": 112,
|
| 1562 |
+
"effective_num_docs": 112,
|
| 1563 |
+
"trust_dataset": true
|
| 1564 |
+
},
|
| 1565 |
+
"leaderboard|mmlu:management": {
|
| 1566 |
+
"name": "mmlu:management",
|
| 1567 |
+
"prompt_function": "mmlu_harness",
|
| 1568 |
+
"hf_repo": "lighteval/mmlu",
|
| 1569 |
+
"hf_subset": "management",
|
| 1570 |
+
"metric": [
|
| 1571 |
+
"loglikelihood_acc"
|
| 1572 |
+
],
|
| 1573 |
+
"hf_avail_splits": [
|
| 1574 |
+
"auxiliary_train",
|
| 1575 |
+
"test",
|
| 1576 |
+
"validation",
|
| 1577 |
+
"dev"
|
| 1578 |
+
],
|
| 1579 |
+
"evaluation_splits": [
|
| 1580 |
+
"test"
|
| 1581 |
+
],
|
| 1582 |
+
"few_shots_split": "dev",
|
| 1583 |
+
"few_shots_select": "sequential",
|
| 1584 |
+
"generation_size": 1,
|
| 1585 |
+
"stop_sequence": [
|
| 1586 |
+
"\n"
|
| 1587 |
+
],
|
| 1588 |
+
"output_regex": null,
|
| 1589 |
+
"frozen": false,
|
| 1590 |
+
"suite": [
|
| 1591 |
+
"leaderboard",
|
| 1592 |
+
"mmlu"
|
| 1593 |
+
],
|
| 1594 |
+
"original_num_docs": 103,
|
| 1595 |
+
"effective_num_docs": 103,
|
| 1596 |
+
"trust_dataset": true
|
| 1597 |
+
},
|
| 1598 |
+
"leaderboard|mmlu:marketing": {
|
| 1599 |
+
"name": "mmlu:marketing",
|
| 1600 |
+
"prompt_function": "mmlu_harness",
|
| 1601 |
+
"hf_repo": "lighteval/mmlu",
|
| 1602 |
+
"hf_subset": "marketing",
|
| 1603 |
+
"metric": [
|
| 1604 |
+
"loglikelihood_acc"
|
| 1605 |
+
],
|
| 1606 |
+
"hf_avail_splits": [
|
| 1607 |
+
"auxiliary_train",
|
| 1608 |
+
"test",
|
| 1609 |
+
"validation",
|
| 1610 |
+
"dev"
|
| 1611 |
+
],
|
| 1612 |
+
"evaluation_splits": [
|
| 1613 |
+
"test"
|
| 1614 |
+
],
|
| 1615 |
+
"few_shots_split": "dev",
|
| 1616 |
+
"few_shots_select": "sequential",
|
| 1617 |
+
"generation_size": 1,
|
| 1618 |
+
"stop_sequence": [
|
| 1619 |
+
"\n"
|
| 1620 |
+
],
|
| 1621 |
+
"output_regex": null,
|
| 1622 |
+
"frozen": false,
|
| 1623 |
+
"suite": [
|
| 1624 |
+
"leaderboard",
|
| 1625 |
+
"mmlu"
|
| 1626 |
+
],
|
| 1627 |
+
"original_num_docs": 234,
|
| 1628 |
+
"effective_num_docs": 234,
|
| 1629 |
+
"trust_dataset": true
|
| 1630 |
+
},
|
| 1631 |
+
"leaderboard|mmlu:medical_genetics": {
|
| 1632 |
+
"name": "mmlu:medical_genetics",
|
| 1633 |
+
"prompt_function": "mmlu_harness",
|
| 1634 |
+
"hf_repo": "lighteval/mmlu",
|
| 1635 |
+
"hf_subset": "medical_genetics",
|
| 1636 |
+
"metric": [
|
| 1637 |
+
"loglikelihood_acc"
|
| 1638 |
+
],
|
| 1639 |
+
"hf_avail_splits": [
|
| 1640 |
+
"auxiliary_train",
|
| 1641 |
+
"test",
|
| 1642 |
+
"validation",
|
| 1643 |
+
"dev"
|
| 1644 |
+
],
|
| 1645 |
+
"evaluation_splits": [
|
| 1646 |
+
"test"
|
| 1647 |
+
],
|
| 1648 |
+
"few_shots_split": "dev",
|
| 1649 |
+
"few_shots_select": "sequential",
|
| 1650 |
+
"generation_size": 1,
|
| 1651 |
+
"stop_sequence": [
|
| 1652 |
+
"\n"
|
| 1653 |
+
],
|
| 1654 |
+
"output_regex": null,
|
| 1655 |
+
"frozen": false,
|
| 1656 |
+
"suite": [
|
| 1657 |
+
"leaderboard",
|
| 1658 |
+
"mmlu"
|
| 1659 |
+
],
|
| 1660 |
+
"original_num_docs": 100,
|
| 1661 |
+
"effective_num_docs": 100,
|
| 1662 |
+
"trust_dataset": true
|
| 1663 |
+
},
|
| 1664 |
+
"leaderboard|mmlu:miscellaneous": {
|
| 1665 |
+
"name": "mmlu:miscellaneous",
|
| 1666 |
+
"prompt_function": "mmlu_harness",
|
| 1667 |
+
"hf_repo": "lighteval/mmlu",
|
| 1668 |
+
"hf_subset": "miscellaneous",
|
| 1669 |
+
"metric": [
|
| 1670 |
+
"loglikelihood_acc"
|
| 1671 |
+
],
|
| 1672 |
+
"hf_avail_splits": [
|
| 1673 |
+
"auxiliary_train",
|
| 1674 |
+
"test",
|
| 1675 |
+
"validation",
|
| 1676 |
+
"dev"
|
| 1677 |
+
],
|
| 1678 |
+
"evaluation_splits": [
|
| 1679 |
+
"test"
|
| 1680 |
+
],
|
| 1681 |
+
"few_shots_split": "dev",
|
| 1682 |
+
"few_shots_select": "sequential",
|
| 1683 |
+
"generation_size": 1,
|
| 1684 |
+
"stop_sequence": [
|
| 1685 |
+
"\n"
|
| 1686 |
+
],
|
| 1687 |
+
"output_regex": null,
|
| 1688 |
+
"frozen": false,
|
| 1689 |
+
"suite": [
|
| 1690 |
+
"leaderboard",
|
| 1691 |
+
"mmlu"
|
| 1692 |
+
],
|
| 1693 |
+
"original_num_docs": 783,
|
| 1694 |
+
"effective_num_docs": 783,
|
| 1695 |
+
"trust_dataset": true
|
| 1696 |
+
},
|
| 1697 |
+
"leaderboard|mmlu:moral_disputes": {
|
| 1698 |
+
"name": "mmlu:moral_disputes",
|
| 1699 |
+
"prompt_function": "mmlu_harness",
|
| 1700 |
+
"hf_repo": "lighteval/mmlu",
|
| 1701 |
+
"hf_subset": "moral_disputes",
|
| 1702 |
+
"metric": [
|
| 1703 |
+
"loglikelihood_acc"
|
| 1704 |
+
],
|
| 1705 |
+
"hf_avail_splits": [
|
| 1706 |
+
"auxiliary_train",
|
| 1707 |
+
"test",
|
| 1708 |
+
"validation",
|
| 1709 |
+
"dev"
|
| 1710 |
+
],
|
| 1711 |
+
"evaluation_splits": [
|
| 1712 |
+
"test"
|
| 1713 |
+
],
|
| 1714 |
+
"few_shots_split": "dev",
|
| 1715 |
+
"few_shots_select": "sequential",
|
| 1716 |
+
"generation_size": 1,
|
| 1717 |
+
"stop_sequence": [
|
| 1718 |
+
"\n"
|
| 1719 |
+
],
|
| 1720 |
+
"output_regex": null,
|
| 1721 |
+
"frozen": false,
|
| 1722 |
+
"suite": [
|
| 1723 |
+
"leaderboard",
|
| 1724 |
+
"mmlu"
|
| 1725 |
+
],
|
| 1726 |
+
"original_num_docs": 346,
|
| 1727 |
+
"effective_num_docs": 346,
|
| 1728 |
+
"trust_dataset": true
|
| 1729 |
+
},
|
| 1730 |
+
"leaderboard|mmlu:moral_scenarios": {
|
| 1731 |
+
"name": "mmlu:moral_scenarios",
|
| 1732 |
+
"prompt_function": "mmlu_harness",
|
| 1733 |
+
"hf_repo": "lighteval/mmlu",
|
| 1734 |
+
"hf_subset": "moral_scenarios",
|
| 1735 |
+
"metric": [
|
| 1736 |
+
"loglikelihood_acc"
|
| 1737 |
+
],
|
| 1738 |
+
"hf_avail_splits": [
|
| 1739 |
+
"auxiliary_train",
|
| 1740 |
+
"test",
|
| 1741 |
+
"validation",
|
| 1742 |
+
"dev"
|
| 1743 |
+
],
|
| 1744 |
+
"evaluation_splits": [
|
| 1745 |
+
"test"
|
| 1746 |
+
],
|
| 1747 |
+
"few_shots_split": "dev",
|
| 1748 |
+
"few_shots_select": "sequential",
|
| 1749 |
+
"generation_size": 1,
|
| 1750 |
+
"stop_sequence": [
|
| 1751 |
+
"\n"
|
| 1752 |
+
],
|
| 1753 |
+
"output_regex": null,
|
| 1754 |
+
"frozen": false,
|
| 1755 |
+
"suite": [
|
| 1756 |
+
"leaderboard",
|
| 1757 |
+
"mmlu"
|
| 1758 |
+
],
|
| 1759 |
+
"original_num_docs": 895,
|
| 1760 |
+
"effective_num_docs": 895,
|
| 1761 |
+
"trust_dataset": true
|
| 1762 |
+
},
|
| 1763 |
+
"leaderboard|mmlu:nutrition": {
|
| 1764 |
+
"name": "mmlu:nutrition",
|
| 1765 |
+
"prompt_function": "mmlu_harness",
|
| 1766 |
+
"hf_repo": "lighteval/mmlu",
|
| 1767 |
+
"hf_subset": "nutrition",
|
| 1768 |
+
"metric": [
|
| 1769 |
+
"loglikelihood_acc"
|
| 1770 |
+
],
|
| 1771 |
+
"hf_avail_splits": [
|
| 1772 |
+
"auxiliary_train",
|
| 1773 |
+
"test",
|
| 1774 |
+
"validation",
|
| 1775 |
+
"dev"
|
| 1776 |
+
],
|
| 1777 |
+
"evaluation_splits": [
|
| 1778 |
+
"test"
|
| 1779 |
+
],
|
| 1780 |
+
"few_shots_split": "dev",
|
| 1781 |
+
"few_shots_select": "sequential",
|
| 1782 |
+
"generation_size": 1,
|
| 1783 |
+
"stop_sequence": [
|
| 1784 |
+
"\n"
|
| 1785 |
+
],
|
| 1786 |
+
"output_regex": null,
|
| 1787 |
+
"frozen": false,
|
| 1788 |
+
"suite": [
|
| 1789 |
+
"leaderboard",
|
| 1790 |
+
"mmlu"
|
| 1791 |
+
],
|
| 1792 |
+
"original_num_docs": 306,
|
| 1793 |
+
"effective_num_docs": 306,
|
| 1794 |
+
"trust_dataset": true
|
| 1795 |
+
},
|
| 1796 |
+
"leaderboard|mmlu:philosophy": {
|
| 1797 |
+
"name": "mmlu:philosophy",
|
| 1798 |
+
"prompt_function": "mmlu_harness",
|
| 1799 |
+
"hf_repo": "lighteval/mmlu",
|
| 1800 |
+
"hf_subset": "philosophy",
|
| 1801 |
+
"metric": [
|
| 1802 |
+
"loglikelihood_acc"
|
| 1803 |
+
],
|
| 1804 |
+
"hf_avail_splits": [
|
| 1805 |
+
"auxiliary_train",
|
| 1806 |
+
"test",
|
| 1807 |
+
"validation",
|
| 1808 |
+
"dev"
|
| 1809 |
+
],
|
| 1810 |
+
"evaluation_splits": [
|
| 1811 |
+
"test"
|
| 1812 |
+
],
|
| 1813 |
+
"few_shots_split": "dev",
|
| 1814 |
+
"few_shots_select": "sequential",
|
| 1815 |
+
"generation_size": 1,
|
| 1816 |
+
"stop_sequence": [
|
| 1817 |
+
"\n"
|
| 1818 |
+
],
|
| 1819 |
+
"output_regex": null,
|
| 1820 |
+
"frozen": false,
|
| 1821 |
+
"suite": [
|
| 1822 |
+
"leaderboard",
|
| 1823 |
+
"mmlu"
|
| 1824 |
+
],
|
| 1825 |
+
"original_num_docs": 311,
|
| 1826 |
+
"effective_num_docs": 311,
|
| 1827 |
+
"trust_dataset": true
|
| 1828 |
+
},
|
| 1829 |
+
"leaderboard|mmlu:prehistory": {
|
| 1830 |
+
"name": "mmlu:prehistory",
|
| 1831 |
+
"prompt_function": "mmlu_harness",
|
| 1832 |
+
"hf_repo": "lighteval/mmlu",
|
| 1833 |
+
"hf_subset": "prehistory",
|
| 1834 |
+
"metric": [
|
| 1835 |
+
"loglikelihood_acc"
|
| 1836 |
+
],
|
| 1837 |
+
"hf_avail_splits": [
|
| 1838 |
+
"auxiliary_train",
|
| 1839 |
+
"test",
|
| 1840 |
+
"validation",
|
| 1841 |
+
"dev"
|
| 1842 |
+
],
|
| 1843 |
+
"evaluation_splits": [
|
| 1844 |
+
"test"
|
| 1845 |
+
],
|
| 1846 |
+
"few_shots_split": "dev",
|
| 1847 |
+
"few_shots_select": "sequential",
|
| 1848 |
+
"generation_size": 1,
|
| 1849 |
+
"stop_sequence": [
|
| 1850 |
+
"\n"
|
| 1851 |
+
],
|
| 1852 |
+
"output_regex": null,
|
| 1853 |
+
"frozen": false,
|
| 1854 |
+
"suite": [
|
| 1855 |
+
"leaderboard",
|
| 1856 |
+
"mmlu"
|
| 1857 |
+
],
|
| 1858 |
+
"original_num_docs": 324,
|
| 1859 |
+
"effective_num_docs": 324,
|
| 1860 |
+
"trust_dataset": true
|
| 1861 |
+
},
|
| 1862 |
+
"leaderboard|mmlu:professional_accounting": {
|
| 1863 |
+
"name": "mmlu:professional_accounting",
|
| 1864 |
+
"prompt_function": "mmlu_harness",
|
| 1865 |
+
"hf_repo": "lighteval/mmlu",
|
| 1866 |
+
"hf_subset": "professional_accounting",
|
| 1867 |
+
"metric": [
|
| 1868 |
+
"loglikelihood_acc"
|
| 1869 |
+
],
|
| 1870 |
+
"hf_avail_splits": [
|
| 1871 |
+
"auxiliary_train",
|
| 1872 |
+
"test",
|
| 1873 |
+
"validation",
|
| 1874 |
+
"dev"
|
| 1875 |
+
],
|
| 1876 |
+
"evaluation_splits": [
|
| 1877 |
+
"test"
|
| 1878 |
+
],
|
| 1879 |
+
"few_shots_split": "dev",
|
| 1880 |
+
"few_shots_select": "sequential",
|
| 1881 |
+
"generation_size": 1,
|
| 1882 |
+
"stop_sequence": [
|
| 1883 |
+
"\n"
|
| 1884 |
+
],
|
| 1885 |
+
"output_regex": null,
|
| 1886 |
+
"frozen": false,
|
| 1887 |
+
"suite": [
|
| 1888 |
+
"leaderboard",
|
| 1889 |
+
"mmlu"
|
| 1890 |
+
],
|
| 1891 |
+
"original_num_docs": 282,
|
| 1892 |
+
"effective_num_docs": 282,
|
| 1893 |
+
"trust_dataset": true
|
| 1894 |
+
},
|
| 1895 |
+
"leaderboard|mmlu:professional_law": {
|
| 1896 |
+
"name": "mmlu:professional_law",
|
| 1897 |
+
"prompt_function": "mmlu_harness",
|
| 1898 |
+
"hf_repo": "lighteval/mmlu",
|
| 1899 |
+
"hf_subset": "professional_law",
|
| 1900 |
+
"metric": [
|
| 1901 |
+
"loglikelihood_acc"
|
| 1902 |
+
],
|
| 1903 |
+
"hf_avail_splits": [
|
| 1904 |
+
"auxiliary_train",
|
| 1905 |
+
"test",
|
| 1906 |
+
"validation",
|
| 1907 |
+
"dev"
|
| 1908 |
+
],
|
| 1909 |
+
"evaluation_splits": [
|
| 1910 |
+
"test"
|
| 1911 |
+
],
|
| 1912 |
+
"few_shots_split": "dev",
|
| 1913 |
+
"few_shots_select": "sequential",
|
| 1914 |
+
"generation_size": 1,
|
| 1915 |
+
"stop_sequence": [
|
| 1916 |
+
"\n"
|
| 1917 |
+
],
|
| 1918 |
+
"output_regex": null,
|
| 1919 |
+
"frozen": false,
|
| 1920 |
+
"suite": [
|
| 1921 |
+
"leaderboard",
|
| 1922 |
+
"mmlu"
|
| 1923 |
+
],
|
| 1924 |
+
"original_num_docs": 1534,
|
| 1925 |
+
"effective_num_docs": 1534,
|
| 1926 |
+
"trust_dataset": true
|
| 1927 |
+
},
|
| 1928 |
+
"leaderboard|mmlu:professional_medicine": {
|
| 1929 |
+
"name": "mmlu:professional_medicine",
|
| 1930 |
+
"prompt_function": "mmlu_harness",
|
| 1931 |
+
"hf_repo": "lighteval/mmlu",
|
| 1932 |
+
"hf_subset": "professional_medicine",
|
| 1933 |
+
"metric": [
|
| 1934 |
+
"loglikelihood_acc"
|
| 1935 |
+
],
|
| 1936 |
+
"hf_avail_splits": [
|
| 1937 |
+
"auxiliary_train",
|
| 1938 |
+
"test",
|
| 1939 |
+
"validation",
|
| 1940 |
+
"dev"
|
| 1941 |
+
],
|
| 1942 |
+
"evaluation_splits": [
|
| 1943 |
+
"test"
|
| 1944 |
+
],
|
| 1945 |
+
"few_shots_split": "dev",
|
| 1946 |
+
"few_shots_select": "sequential",
|
| 1947 |
+
"generation_size": 1,
|
| 1948 |
+
"stop_sequence": [
|
| 1949 |
+
"\n"
|
| 1950 |
+
],
|
| 1951 |
+
"output_regex": null,
|
| 1952 |
+
"frozen": false,
|
| 1953 |
+
"suite": [
|
| 1954 |
+
"leaderboard",
|
| 1955 |
+
"mmlu"
|
| 1956 |
+
],
|
| 1957 |
+
"original_num_docs": 272,
|
| 1958 |
+
"effective_num_docs": 272,
|
| 1959 |
+
"trust_dataset": true
|
| 1960 |
+
},
|
| 1961 |
+
"leaderboard|mmlu:professional_psychology": {
|
| 1962 |
+
"name": "mmlu:professional_psychology",
|
| 1963 |
+
"prompt_function": "mmlu_harness",
|
| 1964 |
+
"hf_repo": "lighteval/mmlu",
|
| 1965 |
+
"hf_subset": "professional_psychology",
|
| 1966 |
+
"metric": [
|
| 1967 |
+
"loglikelihood_acc"
|
| 1968 |
+
],
|
| 1969 |
+
"hf_avail_splits": [
|
| 1970 |
+
"auxiliary_train",
|
| 1971 |
+
"test",
|
| 1972 |
+
"validation",
|
| 1973 |
+
"dev"
|
| 1974 |
+
],
|
| 1975 |
+
"evaluation_splits": [
|
| 1976 |
+
"test"
|
| 1977 |
+
],
|
| 1978 |
+
"few_shots_split": "dev",
|
| 1979 |
+
"few_shots_select": "sequential",
|
| 1980 |
+
"generation_size": 1,
|
| 1981 |
+
"stop_sequence": [
|
| 1982 |
+
"\n"
|
| 1983 |
+
],
|
| 1984 |
+
"output_regex": null,
|
| 1985 |
+
"frozen": false,
|
| 1986 |
+
"suite": [
|
| 1987 |
+
"leaderboard",
|
| 1988 |
+
"mmlu"
|
| 1989 |
+
],
|
| 1990 |
+
"original_num_docs": 612,
|
| 1991 |
+
"effective_num_docs": 612,
|
| 1992 |
+
"trust_dataset": true
|
| 1993 |
+
},
|
| 1994 |
+
"leaderboard|mmlu:public_relations": {
|
| 1995 |
+
"name": "mmlu:public_relations",
|
| 1996 |
+
"prompt_function": "mmlu_harness",
|
| 1997 |
+
"hf_repo": "lighteval/mmlu",
|
| 1998 |
+
"hf_subset": "public_relations",
|
| 1999 |
+
"metric": [
|
| 2000 |
+
"loglikelihood_acc"
|
| 2001 |
+
],
|
| 2002 |
+
"hf_avail_splits": [
|
| 2003 |
+
"auxiliary_train",
|
| 2004 |
+
"test",
|
| 2005 |
+
"validation",
|
| 2006 |
+
"dev"
|
| 2007 |
+
],
|
| 2008 |
+
"evaluation_splits": [
|
| 2009 |
+
"test"
|
| 2010 |
+
],
|
| 2011 |
+
"few_shots_split": "dev",
|
| 2012 |
+
"few_shots_select": "sequential",
|
| 2013 |
+
"generation_size": 1,
|
| 2014 |
+
"stop_sequence": [
|
| 2015 |
+
"\n"
|
| 2016 |
+
],
|
| 2017 |
+
"output_regex": null,
|
| 2018 |
+
"frozen": false,
|
| 2019 |
+
"suite": [
|
| 2020 |
+
"leaderboard",
|
| 2021 |
+
"mmlu"
|
| 2022 |
+
],
|
| 2023 |
+
"original_num_docs": 110,
|
| 2024 |
+
"effective_num_docs": 110,
|
| 2025 |
+
"trust_dataset": true
|
| 2026 |
+
},
|
| 2027 |
+
"leaderboard|mmlu:security_studies": {
|
| 2028 |
+
"name": "mmlu:security_studies",
|
| 2029 |
+
"prompt_function": "mmlu_harness",
|
| 2030 |
+
"hf_repo": "lighteval/mmlu",
|
| 2031 |
+
"hf_subset": "security_studies",
|
| 2032 |
+
"metric": [
|
| 2033 |
+
"loglikelihood_acc"
|
| 2034 |
+
],
|
| 2035 |
+
"hf_avail_splits": [
|
| 2036 |
+
"auxiliary_train",
|
| 2037 |
+
"test",
|
| 2038 |
+
"validation",
|
| 2039 |
+
"dev"
|
| 2040 |
+
],
|
| 2041 |
+
"evaluation_splits": [
|
| 2042 |
+
"test"
|
| 2043 |
+
],
|
| 2044 |
+
"few_shots_split": "dev",
|
| 2045 |
+
"few_shots_select": "sequential",
|
| 2046 |
+
"generation_size": 1,
|
| 2047 |
+
"stop_sequence": [
|
| 2048 |
+
"\n"
|
| 2049 |
+
],
|
| 2050 |
+
"output_regex": null,
|
| 2051 |
+
"frozen": false,
|
| 2052 |
+
"suite": [
|
| 2053 |
+
"leaderboard",
|
| 2054 |
+
"mmlu"
|
| 2055 |
+
],
|
| 2056 |
+
"original_num_docs": 245,
|
| 2057 |
+
"effective_num_docs": 245,
|
| 2058 |
+
"trust_dataset": true
|
| 2059 |
+
},
|
| 2060 |
+
"leaderboard|mmlu:sociology": {
|
| 2061 |
+
"name": "mmlu:sociology",
|
| 2062 |
+
"prompt_function": "mmlu_harness",
|
| 2063 |
+
"hf_repo": "lighteval/mmlu",
|
| 2064 |
+
"hf_subset": "sociology",
|
| 2065 |
+
"metric": [
|
| 2066 |
+
"loglikelihood_acc"
|
| 2067 |
+
],
|
| 2068 |
+
"hf_avail_splits": [
|
| 2069 |
+
"auxiliary_train",
|
| 2070 |
+
"test",
|
| 2071 |
+
"validation",
|
| 2072 |
+
"dev"
|
| 2073 |
+
],
|
| 2074 |
+
"evaluation_splits": [
|
| 2075 |
+
"test"
|
| 2076 |
+
],
|
| 2077 |
+
"few_shots_split": "dev",
|
| 2078 |
+
"few_shots_select": "sequential",
|
| 2079 |
+
"generation_size": 1,
|
| 2080 |
+
"stop_sequence": [
|
| 2081 |
+
"\n"
|
| 2082 |
+
],
|
| 2083 |
+
"output_regex": null,
|
| 2084 |
+
"frozen": false,
|
| 2085 |
+
"suite": [
|
| 2086 |
+
"leaderboard",
|
| 2087 |
+
"mmlu"
|
| 2088 |
+
],
|
| 2089 |
+
"original_num_docs": 201,
|
| 2090 |
+
"effective_num_docs": 201,
|
| 2091 |
+
"trust_dataset": true
|
| 2092 |
+
},
|
| 2093 |
+
"leaderboard|mmlu:us_foreign_policy": {
|
| 2094 |
+
"name": "mmlu:us_foreign_policy",
|
| 2095 |
+
"prompt_function": "mmlu_harness",
|
| 2096 |
+
"hf_repo": "lighteval/mmlu",
|
| 2097 |
+
"hf_subset": "us_foreign_policy",
|
| 2098 |
+
"metric": [
|
| 2099 |
+
"loglikelihood_acc"
|
| 2100 |
+
],
|
| 2101 |
+
"hf_avail_splits": [
|
| 2102 |
+
"auxiliary_train",
|
| 2103 |
+
"test",
|
| 2104 |
+
"validation",
|
| 2105 |
+
"dev"
|
| 2106 |
+
],
|
| 2107 |
+
"evaluation_splits": [
|
| 2108 |
+
"test"
|
| 2109 |
+
],
|
| 2110 |
+
"few_shots_split": "dev",
|
| 2111 |
+
"few_shots_select": "sequential",
|
| 2112 |
+
"generation_size": 1,
|
| 2113 |
+
"stop_sequence": [
|
| 2114 |
+
"\n"
|
| 2115 |
+
],
|
| 2116 |
+
"output_regex": null,
|
| 2117 |
+
"frozen": false,
|
| 2118 |
+
"suite": [
|
| 2119 |
+
"leaderboard",
|
| 2120 |
+
"mmlu"
|
| 2121 |
+
],
|
| 2122 |
+
"original_num_docs": 100,
|
| 2123 |
+
"effective_num_docs": 100,
|
| 2124 |
+
"trust_dataset": true
|
| 2125 |
+
},
|
| 2126 |
+
"leaderboard|mmlu:virology": {
|
| 2127 |
+
"name": "mmlu:virology",
|
| 2128 |
+
"prompt_function": "mmlu_harness",
|
| 2129 |
+
"hf_repo": "lighteval/mmlu",
|
| 2130 |
+
"hf_subset": "virology",
|
| 2131 |
+
"metric": [
|
| 2132 |
+
"loglikelihood_acc"
|
| 2133 |
+
],
|
| 2134 |
+
"hf_avail_splits": [
|
| 2135 |
+
"auxiliary_train",
|
| 2136 |
+
"test",
|
| 2137 |
+
"validation",
|
| 2138 |
+
"dev"
|
| 2139 |
+
],
|
| 2140 |
+
"evaluation_splits": [
|
| 2141 |
+
"test"
|
| 2142 |
+
],
|
| 2143 |
+
"few_shots_split": "dev",
|
| 2144 |
+
"few_shots_select": "sequential",
|
| 2145 |
+
"generation_size": 1,
|
| 2146 |
+
"stop_sequence": [
|
| 2147 |
+
"\n"
|
| 2148 |
+
],
|
| 2149 |
+
"output_regex": null,
|
| 2150 |
+
"frozen": false,
|
| 2151 |
+
"suite": [
|
| 2152 |
+
"leaderboard",
|
| 2153 |
+
"mmlu"
|
| 2154 |
+
],
|
| 2155 |
+
"original_num_docs": 166,
|
| 2156 |
+
"effective_num_docs": 166,
|
| 2157 |
+
"trust_dataset": true
|
| 2158 |
+
},
|
| 2159 |
+
"leaderboard|mmlu:world_religions": {
|
| 2160 |
+
"name": "mmlu:world_religions",
|
| 2161 |
+
"prompt_function": "mmlu_harness",
|
| 2162 |
+
"hf_repo": "lighteval/mmlu",
|
| 2163 |
+
"hf_subset": "world_religions",
|
| 2164 |
+
"metric": [
|
| 2165 |
+
"loglikelihood_acc"
|
| 2166 |
+
],
|
| 2167 |
+
"hf_avail_splits": [
|
| 2168 |
+
"auxiliary_train",
|
| 2169 |
+
"test",
|
| 2170 |
+
"validation",
|
| 2171 |
+
"dev"
|
| 2172 |
+
],
|
| 2173 |
+
"evaluation_splits": [
|
| 2174 |
+
"test"
|
| 2175 |
+
],
|
| 2176 |
+
"few_shots_split": "dev",
|
| 2177 |
+
"few_shots_select": "sequential",
|
| 2178 |
+
"generation_size": 1,
|
| 2179 |
+
"stop_sequence": [
|
| 2180 |
+
"\n"
|
| 2181 |
+
],
|
| 2182 |
+
"output_regex": null,
|
| 2183 |
+
"frozen": false,
|
| 2184 |
+
"suite": [
|
| 2185 |
+
"leaderboard",
|
| 2186 |
+
"mmlu"
|
| 2187 |
+
],
|
| 2188 |
+
"original_num_docs": 171,
|
| 2189 |
+
"effective_num_docs": 171,
|
| 2190 |
+
"trust_dataset": true
|
| 2191 |
+
}
|
| 2192 |
+
},
|
| 2193 |
+
"summary_tasks": {
|
| 2194 |
+
"leaderboard|mmlu:abstract_algebra|5": {
|
| 2195 |
+
"hashes": {
|
| 2196 |
+
"hash_examples": "4c76229e00c9c0e9",
|
| 2197 |
+
"hash_full_prompts": "a45d01c3409c889c",
|
| 2198 |
+
"hash_input_tokens": "d0571b6ffb835507",
|
| 2199 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2200 |
+
},
|
| 2201 |
+
"truncated": 0,
|
| 2202 |
+
"non_truncated": 100,
|
| 2203 |
+
"padded": 400,
|
| 2204 |
+
"non_padded": 0,
|
| 2205 |
+
"effective_few_shots": 5.0,
|
| 2206 |
+
"num_truncated_few_shots": 0
|
| 2207 |
+
},
|
| 2208 |
+
"leaderboard|mmlu:anatomy|5": {
|
| 2209 |
+
"hashes": {
|
| 2210 |
+
"hash_examples": "6a1f8104dccbd33b",
|
| 2211 |
+
"hash_full_prompts": "e245c6600e03cc32",
|
| 2212 |
+
"hash_input_tokens": "8dd20ec55e9ad889",
|
| 2213 |
+
"hash_cont_tokens": "263324e6ce7f9b36"
|
| 2214 |
+
},
|
| 2215 |
+
"truncated": 0,
|
| 2216 |
+
"non_truncated": 135,
|
| 2217 |
+
"padded": 540,
|
| 2218 |
+
"non_padded": 0,
|
| 2219 |
+
"effective_few_shots": 5.0,
|
| 2220 |
+
"num_truncated_few_shots": 0
|
| 2221 |
+
},
|
| 2222 |
+
"leaderboard|mmlu:astronomy|5": {
|
| 2223 |
+
"hashes": {
|
| 2224 |
+
"hash_examples": "1302effa3a76ce4c",
|
| 2225 |
+
"hash_full_prompts": "390f9bddf857ad04",
|
| 2226 |
+
"hash_input_tokens": "81e8167c0c820f24",
|
| 2227 |
+
"hash_cont_tokens": "18ba399c6801138e"
|
| 2228 |
+
},
|
| 2229 |
+
"truncated": 0,
|
| 2230 |
+
"non_truncated": 152,
|
| 2231 |
+
"padded": 608,
|
| 2232 |
+
"non_padded": 0,
|
| 2233 |
+
"effective_few_shots": 5.0,
|
| 2234 |
+
"num_truncated_few_shots": 0
|
| 2235 |
+
},
|
| 2236 |
+
"leaderboard|mmlu:business_ethics|5": {
|
| 2237 |
+
"hashes": {
|
| 2238 |
+
"hash_examples": "03cb8bce5336419a",
|
| 2239 |
+
"hash_full_prompts": "5504f893bc4f2fa1",
|
| 2240 |
+
"hash_input_tokens": "668443aa86633b73",
|
| 2241 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2242 |
+
},
|
| 2243 |
+
"truncated": 0,
|
| 2244 |
+
"non_truncated": 100,
|
| 2245 |
+
"padded": 400,
|
| 2246 |
+
"non_padded": 0,
|
| 2247 |
+
"effective_few_shots": 5.0,
|
| 2248 |
+
"num_truncated_few_shots": 0
|
| 2249 |
+
},
|
| 2250 |
+
"leaderboard|mmlu:clinical_knowledge|5": {
|
| 2251 |
+
"hashes": {
|
| 2252 |
+
"hash_examples": "ffbb9c7b2be257f9",
|
| 2253 |
+
"hash_full_prompts": "106ad0bab4b90b78",
|
| 2254 |
+
"hash_input_tokens": "726c176b444e3c55",
|
| 2255 |
+
"hash_cont_tokens": "9d7500060e0dd995"
|
| 2256 |
+
},
|
| 2257 |
+
"truncated": 0,
|
| 2258 |
+
"non_truncated": 265,
|
| 2259 |
+
"padded": 1060,
|
| 2260 |
+
"non_padded": 0,
|
| 2261 |
+
"effective_few_shots": 5.0,
|
| 2262 |
+
"num_truncated_few_shots": 0
|
| 2263 |
+
},
|
| 2264 |
+
"leaderboard|mmlu:college_biology|5": {
|
| 2265 |
+
"hashes": {
|
| 2266 |
+
"hash_examples": "3ee77f176f38eb8e",
|
| 2267 |
+
"hash_full_prompts": "59f9bdf2695cb226",
|
| 2268 |
+
"hash_input_tokens": "7535ef44daca8b2e",
|
| 2269 |
+
"hash_cont_tokens": "78a731af5d2f6472"
|
| 2270 |
+
},
|
| 2271 |
+
"truncated": 0,
|
| 2272 |
+
"non_truncated": 144,
|
| 2273 |
+
"padded": 576,
|
| 2274 |
+
"non_padded": 0,
|
| 2275 |
+
"effective_few_shots": 5.0,
|
| 2276 |
+
"num_truncated_few_shots": 0
|
| 2277 |
+
},
|
| 2278 |
+
"leaderboard|mmlu:college_chemistry|5": {
|
| 2279 |
+
"hashes": {
|
| 2280 |
+
"hash_examples": "ce61a69c46d47aeb",
|
| 2281 |
+
"hash_full_prompts": "3cac9b759fcff7a0",
|
| 2282 |
+
"hash_input_tokens": "e98bdaf1fa27ef3b",
|
| 2283 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2284 |
+
},
|
| 2285 |
+
"truncated": 0,
|
| 2286 |
+
"non_truncated": 100,
|
| 2287 |
+
"padded": 400,
|
| 2288 |
+
"non_padded": 0,
|
| 2289 |
+
"effective_few_shots": 5.0,
|
| 2290 |
+
"num_truncated_few_shots": 0
|
| 2291 |
+
},
|
| 2292 |
+
"leaderboard|mmlu:college_computer_science|5": {
|
| 2293 |
+
"hashes": {
|
| 2294 |
+
"hash_examples": "32805b52d7d5daab",
|
| 2295 |
+
"hash_full_prompts": "010b0cca35070130",
|
| 2296 |
+
"hash_input_tokens": "40494a193cf906d1",
|
| 2297 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2298 |
+
},
|
| 2299 |
+
"truncated": 0,
|
| 2300 |
+
"non_truncated": 100,
|
| 2301 |
+
"padded": 400,
|
| 2302 |
+
"non_padded": 0,
|
| 2303 |
+
"effective_few_shots": 5.0,
|
| 2304 |
+
"num_truncated_few_shots": 0
|
| 2305 |
+
},
|
| 2306 |
+
"leaderboard|mmlu:college_mathematics|5": {
|
| 2307 |
+
"hashes": {
|
| 2308 |
+
"hash_examples": "55da1a0a0bd33722",
|
| 2309 |
+
"hash_full_prompts": "511422eb9eefc773",
|
| 2310 |
+
"hash_input_tokens": "2f512892d24b0086",
|
| 2311 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2312 |
+
},
|
| 2313 |
+
"truncated": 0,
|
| 2314 |
+
"non_truncated": 100,
|
| 2315 |
+
"padded": 400,
|
| 2316 |
+
"non_padded": 0,
|
| 2317 |
+
"effective_few_shots": 5.0,
|
| 2318 |
+
"num_truncated_few_shots": 0
|
| 2319 |
+
},
|
| 2320 |
+
"leaderboard|mmlu:college_medicine|5": {
|
| 2321 |
+
"hashes": {
|
| 2322 |
+
"hash_examples": "c33e143163049176",
|
| 2323 |
+
"hash_full_prompts": "c8cc1a82a51a046e",
|
| 2324 |
+
"hash_input_tokens": "41ba4385551feaf3",
|
| 2325 |
+
"hash_cont_tokens": "699c8eb24e3e446b"
|
| 2326 |
+
},
|
| 2327 |
+
"truncated": 0,
|
| 2328 |
+
"non_truncated": 173,
|
| 2329 |
+
"padded": 692,
|
| 2330 |
+
"non_padded": 0,
|
| 2331 |
+
"effective_few_shots": 5.0,
|
| 2332 |
+
"num_truncated_few_shots": 0
|
| 2333 |
+
},
|
| 2334 |
+
"leaderboard|mmlu:college_physics|5": {
|
| 2335 |
+
"hashes": {
|
| 2336 |
+
"hash_examples": "ebdab1cdb7e555df",
|
| 2337 |
+
"hash_full_prompts": "e40721b5059c5818",
|
| 2338 |
+
"hash_input_tokens": "1f357d859f4e78c2",
|
| 2339 |
+
"hash_cont_tokens": "075997110cbe055e"
|
| 2340 |
+
},
|
| 2341 |
+
"truncated": 0,
|
| 2342 |
+
"non_truncated": 102,
|
| 2343 |
+
"padded": 408,
|
| 2344 |
+
"non_padded": 0,
|
| 2345 |
+
"effective_few_shots": 5.0,
|
| 2346 |
+
"num_truncated_few_shots": 0
|
| 2347 |
+
},
|
| 2348 |
+
"leaderboard|mmlu:computer_security|5": {
|
| 2349 |
+
"hashes": {
|
| 2350 |
+
"hash_examples": "a24fd7d08a560921",
|
| 2351 |
+
"hash_full_prompts": "946c9be5964ac44a",
|
| 2352 |
+
"hash_input_tokens": "def9fb5a2fab003a",
|
| 2353 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2354 |
+
},
|
| 2355 |
+
"truncated": 0,
|
| 2356 |
+
"non_truncated": 100,
|
| 2357 |
+
"padded": 400,
|
| 2358 |
+
"non_padded": 0,
|
| 2359 |
+
"effective_few_shots": 5.0,
|
| 2360 |
+
"num_truncated_few_shots": 0
|
| 2361 |
+
},
|
| 2362 |
+
"leaderboard|mmlu:conceptual_physics|5": {
|
| 2363 |
+
"hashes": {
|
| 2364 |
+
"hash_examples": "8300977a79386993",
|
| 2365 |
+
"hash_full_prompts": "506a4f6094cc40c9",
|
| 2366 |
+
"hash_input_tokens": "b398cceaff8512f7",
|
| 2367 |
+
"hash_cont_tokens": "f22daa6d4818086f"
|
| 2368 |
+
},
|
| 2369 |
+
"truncated": 0,
|
| 2370 |
+
"non_truncated": 235,
|
| 2371 |
+
"padded": 940,
|
| 2372 |
+
"non_padded": 0,
|
| 2373 |
+
"effective_few_shots": 5.0,
|
| 2374 |
+
"num_truncated_few_shots": 0
|
| 2375 |
+
},
|
| 2376 |
+
"leaderboard|mmlu:econometrics|5": {
|
| 2377 |
+
"hashes": {
|
| 2378 |
+
"hash_examples": "ddde36788a04a46f",
|
| 2379 |
+
"hash_full_prompts": "4ed2703f27f1ed05",
|
| 2380 |
+
"hash_input_tokens": "cf227ca8af4bc815",
|
| 2381 |
+
"hash_cont_tokens": "26791a0b1941b4c4"
|
| 2382 |
+
},
|
| 2383 |
+
"truncated": 0,
|
| 2384 |
+
"non_truncated": 114,
|
| 2385 |
+
"padded": 456,
|
| 2386 |
+
"non_padded": 0,
|
| 2387 |
+
"effective_few_shots": 5.0,
|
| 2388 |
+
"num_truncated_few_shots": 0
|
| 2389 |
+
},
|
| 2390 |
+
"leaderboard|mmlu:electrical_engineering|5": {
|
| 2391 |
+
"hashes": {
|
| 2392 |
+
"hash_examples": "acbc5def98c19b3f",
|
| 2393 |
+
"hash_full_prompts": "d8f4b3e11c23653c",
|
| 2394 |
+
"hash_input_tokens": "295e278cbce7ed04",
|
| 2395 |
+
"hash_cont_tokens": "3e336577994f6c0d"
|
| 2396 |
+
},
|
| 2397 |
+
"truncated": 0,
|
| 2398 |
+
"non_truncated": 145,
|
| 2399 |
+
"padded": 580,
|
| 2400 |
+
"non_padded": 0,
|
| 2401 |
+
"effective_few_shots": 5.0,
|
| 2402 |
+
"num_truncated_few_shots": 0
|
| 2403 |
+
},
|
| 2404 |
+
"leaderboard|mmlu:elementary_mathematics|5": {
|
| 2405 |
+
"hashes": {
|
| 2406 |
+
"hash_examples": "146e61d07497a9bd",
|
| 2407 |
+
"hash_full_prompts": "256d111bd15647ff",
|
| 2408 |
+
"hash_input_tokens": "2474a420d7b931ff",
|
| 2409 |
+
"hash_cont_tokens": "1d6bbfa8a67327c8"
|
| 2410 |
+
},
|
| 2411 |
+
"truncated": 0,
|
| 2412 |
+
"non_truncated": 378,
|
| 2413 |
+
"padded": 1512,
|
| 2414 |
+
"non_padded": 0,
|
| 2415 |
+
"effective_few_shots": 5.0,
|
| 2416 |
+
"num_truncated_few_shots": 0
|
| 2417 |
+
},
|
| 2418 |
+
"leaderboard|mmlu:formal_logic|5": {
|
| 2419 |
+
"hashes": {
|
| 2420 |
+
"hash_examples": "8635216e1909a03f",
|
| 2421 |
+
"hash_full_prompts": "1171d04f3b1a11f5",
|
| 2422 |
+
"hash_input_tokens": "f269941d7dabea05",
|
| 2423 |
+
"hash_cont_tokens": "60508d85eb7693a4"
|
| 2424 |
+
},
|
| 2425 |
+
"truncated": 0,
|
| 2426 |
+
"non_truncated": 126,
|
| 2427 |
+
"padded": 504,
|
| 2428 |
+
"non_padded": 0,
|
| 2429 |
+
"effective_few_shots": 5.0,
|
| 2430 |
+
"num_truncated_few_shots": 0
|
| 2431 |
+
},
|
| 2432 |
+
"leaderboard|mmlu:global_facts|5": {
|
| 2433 |
+
"hashes": {
|
| 2434 |
+
"hash_examples": "30b315aa6353ee47",
|
| 2435 |
+
"hash_full_prompts": "a7e56dbc074c7529",
|
| 2436 |
+
"hash_input_tokens": "2036a912407797e6",
|
| 2437 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2438 |
+
},
|
| 2439 |
+
"truncated": 0,
|
| 2440 |
+
"non_truncated": 100,
|
| 2441 |
+
"padded": 400,
|
| 2442 |
+
"non_padded": 0,
|
| 2443 |
+
"effective_few_shots": 5.0,
|
| 2444 |
+
"num_truncated_few_shots": 0
|
| 2445 |
+
},
|
| 2446 |
+
"leaderboard|mmlu:high_school_biology|5": {
|
| 2447 |
+
"hashes": {
|
| 2448 |
+
"hash_examples": "c9136373af2180de",
|
| 2449 |
+
"hash_full_prompts": "ad6e859ed978e04a",
|
| 2450 |
+
"hash_input_tokens": "1bc8ad087ca8f65b",
|
| 2451 |
+
"hash_cont_tokens": "d236ce982144e65f"
|
| 2452 |
+
},
|
| 2453 |
+
"truncated": 0,
|
| 2454 |
+
"non_truncated": 310,
|
| 2455 |
+
"padded": 1240,
|
| 2456 |
+
"non_padded": 0,
|
| 2457 |
+
"effective_few_shots": 5.0,
|
| 2458 |
+
"num_truncated_few_shots": 0
|
| 2459 |
+
},
|
| 2460 |
+
"leaderboard|mmlu:high_school_chemistry|5": {
|
| 2461 |
+
"hashes": {
|
| 2462 |
+
"hash_examples": "b0661bfa1add6404",
|
| 2463 |
+
"hash_full_prompts": "6eb9c04bcc8a8f2a",
|
| 2464 |
+
"hash_input_tokens": "ead708921e3a1c93",
|
| 2465 |
+
"hash_cont_tokens": "59f93238ec5aead6"
|
| 2466 |
+
},
|
| 2467 |
+
"truncated": 0,
|
| 2468 |
+
"non_truncated": 203,
|
| 2469 |
+
"padded": 812,
|
| 2470 |
+
"non_padded": 0,
|
| 2471 |
+
"effective_few_shots": 5.0,
|
| 2472 |
+
"num_truncated_few_shots": 0
|
| 2473 |
+
},
|
| 2474 |
+
"leaderboard|mmlu:high_school_computer_science|5": {
|
| 2475 |
+
"hashes": {
|
| 2476 |
+
"hash_examples": "80fc1d623a3d665f",
|
| 2477 |
+
"hash_full_prompts": "8e51bc91c81cf8dd",
|
| 2478 |
+
"hash_input_tokens": "604f88a2f17d5159",
|
| 2479 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2480 |
+
},
|
| 2481 |
+
"truncated": 0,
|
| 2482 |
+
"non_truncated": 100,
|
| 2483 |
+
"padded": 400,
|
| 2484 |
+
"non_padded": 0,
|
| 2485 |
+
"effective_few_shots": 5.0,
|
| 2486 |
+
"num_truncated_few_shots": 0
|
| 2487 |
+
},
|
| 2488 |
+
"leaderboard|mmlu:high_school_european_history|5": {
|
| 2489 |
+
"hashes": {
|
| 2490 |
+
"hash_examples": "854da6e5af0fe1a1",
|
| 2491 |
+
"hash_full_prompts": "664a1f16c9f3195c",
|
| 2492 |
+
"hash_input_tokens": "1dfe455312f2e6cf",
|
| 2493 |
+
"hash_cont_tokens": "7b7414d6a5da3d91"
|
| 2494 |
+
},
|
| 2495 |
+
"truncated": 0,
|
| 2496 |
+
"non_truncated": 165,
|
| 2497 |
+
"padded": 656,
|
| 2498 |
+
"non_padded": 4,
|
| 2499 |
+
"effective_few_shots": 5.0,
|
| 2500 |
+
"num_truncated_few_shots": 0
|
| 2501 |
+
},
|
| 2502 |
+
"leaderboard|mmlu:high_school_geography|5": {
|
| 2503 |
+
"hashes": {
|
| 2504 |
+
"hash_examples": "7dc963c7acd19ad8",
|
| 2505 |
+
"hash_full_prompts": "f3acf911f4023c8a",
|
| 2506 |
+
"hash_input_tokens": "1985ba6f69f57d66",
|
| 2507 |
+
"hash_cont_tokens": "1b66289e10988f84"
|
| 2508 |
+
},
|
| 2509 |
+
"truncated": 0,
|
| 2510 |
+
"non_truncated": 198,
|
| 2511 |
+
"padded": 792,
|
| 2512 |
+
"non_padded": 0,
|
| 2513 |
+
"effective_few_shots": 5.0,
|
| 2514 |
+
"num_truncated_few_shots": 0
|
| 2515 |
+
},
|
| 2516 |
+
"leaderboard|mmlu:high_school_government_and_politics|5": {
|
| 2517 |
+
"hashes": {
|
| 2518 |
+
"hash_examples": "1f675dcdebc9758f",
|
| 2519 |
+
"hash_full_prompts": "066254feaa3158ae",
|
| 2520 |
+
"hash_input_tokens": "e6960d7d906ffb15",
|
| 2521 |
+
"hash_cont_tokens": "5ab3c3415b1d3a55"
|
| 2522 |
+
},
|
| 2523 |
+
"truncated": 0,
|
| 2524 |
+
"non_truncated": 193,
|
| 2525 |
+
"padded": 772,
|
| 2526 |
+
"non_padded": 0,
|
| 2527 |
+
"effective_few_shots": 5.0,
|
| 2528 |
+
"num_truncated_few_shots": 0
|
| 2529 |
+
},
|
| 2530 |
+
"leaderboard|mmlu:high_school_macroeconomics|5": {
|
| 2531 |
+
"hashes": {
|
| 2532 |
+
"hash_examples": "2fb32cf2d80f0b35",
|
| 2533 |
+
"hash_full_prompts": "19a7fa502aa85c95",
|
| 2534 |
+
"hash_input_tokens": "4ea59b7b8c4856d2",
|
| 2535 |
+
"hash_cont_tokens": "2f5457058d187374"
|
| 2536 |
+
},
|
| 2537 |
+
"truncated": 0,
|
| 2538 |
+
"non_truncated": 390,
|
| 2539 |
+
"padded": 1557,
|
| 2540 |
+
"non_padded": 3,
|
| 2541 |
+
"effective_few_shots": 5.0,
|
| 2542 |
+
"num_truncated_few_shots": 0
|
| 2543 |
+
},
|
| 2544 |
+
"leaderboard|mmlu:high_school_mathematics|5": {
|
| 2545 |
+
"hashes": {
|
| 2546 |
+
"hash_examples": "fd6646fdb5d58a1f",
|
| 2547 |
+
"hash_full_prompts": "4f704e369778b5b0",
|
| 2548 |
+
"hash_input_tokens": "7d39279726411bb3",
|
| 2549 |
+
"hash_cont_tokens": "e35137cb972e1918"
|
| 2550 |
+
},
|
| 2551 |
+
"truncated": 0,
|
| 2552 |
+
"non_truncated": 270,
|
| 2553 |
+
"padded": 1080,
|
| 2554 |
+
"non_padded": 0,
|
| 2555 |
+
"effective_few_shots": 5.0,
|
| 2556 |
+
"num_truncated_few_shots": 0
|
| 2557 |
+
},
|
| 2558 |
+
"leaderboard|mmlu:high_school_microeconomics|5": {
|
| 2559 |
+
"hashes": {
|
| 2560 |
+
"hash_examples": "2118f21f71d87d84",
|
| 2561 |
+
"hash_full_prompts": "4350f9e2240f8010",
|
| 2562 |
+
"hash_input_tokens": "2be919ac2e73f3d1",
|
| 2563 |
+
"hash_cont_tokens": "f756093278ebb83e"
|
| 2564 |
+
},
|
| 2565 |
+
"truncated": 0,
|
| 2566 |
+
"non_truncated": 238,
|
| 2567 |
+
"padded": 908,
|
| 2568 |
+
"non_padded": 44,
|
| 2569 |
+
"effective_few_shots": 5.0,
|
| 2570 |
+
"num_truncated_few_shots": 0
|
| 2571 |
+
},
|
| 2572 |
+
"leaderboard|mmlu:high_school_physics|5": {
|
| 2573 |
+
"hashes": {
|
| 2574 |
+
"hash_examples": "dc3ce06378548565",
|
| 2575 |
+
"hash_full_prompts": "5dc0d6831b66188f",
|
| 2576 |
+
"hash_input_tokens": "9b2e07d3183ade24",
|
| 2577 |
+
"hash_cont_tokens": "9cf883ebf1c82176"
|
| 2578 |
+
},
|
| 2579 |
+
"truncated": 0,
|
| 2580 |
+
"non_truncated": 151,
|
| 2581 |
+
"padded": 604,
|
| 2582 |
+
"non_padded": 0,
|
| 2583 |
+
"effective_few_shots": 5.0,
|
| 2584 |
+
"num_truncated_few_shots": 0
|
| 2585 |
+
},
|
| 2586 |
+
"leaderboard|mmlu:high_school_psychology|5": {
|
| 2587 |
+
"hashes": {
|
| 2588 |
+
"hash_examples": "c8d1d98a40e11f2f",
|
| 2589 |
+
"hash_full_prompts": "af2b097da6d50365",
|
| 2590 |
+
"hash_input_tokens": "a0f7b561c0177eb7",
|
| 2591 |
+
"hash_cont_tokens": "bda0f77331ebb21a"
|
| 2592 |
+
},
|
| 2593 |
+
"truncated": 0,
|
| 2594 |
+
"non_truncated": 545,
|
| 2595 |
+
"padded": 2178,
|
| 2596 |
+
"non_padded": 2,
|
| 2597 |
+
"effective_few_shots": 5.0,
|
| 2598 |
+
"num_truncated_few_shots": 0
|
| 2599 |
+
},
|
| 2600 |
+
"leaderboard|mmlu:high_school_statistics|5": {
|
| 2601 |
+
"hashes": {
|
| 2602 |
+
"hash_examples": "666c8759b98ee4ff",
|
| 2603 |
+
"hash_full_prompts": "c757694421d6d68d",
|
| 2604 |
+
"hash_input_tokens": "0e353fc06f61e59b",
|
| 2605 |
+
"hash_cont_tokens": "4d04f014105a0bad"
|
| 2606 |
+
},
|
| 2607 |
+
"truncated": 0,
|
| 2608 |
+
"non_truncated": 216,
|
| 2609 |
+
"padded": 864,
|
| 2610 |
+
"non_padded": 0,
|
| 2611 |
+
"effective_few_shots": 5.0,
|
| 2612 |
+
"num_truncated_few_shots": 0
|
| 2613 |
+
},
|
| 2614 |
+
"leaderboard|mmlu:high_school_us_history|5": {
|
| 2615 |
+
"hashes": {
|
| 2616 |
+
"hash_examples": "95fef1c4b7d3f81e",
|
| 2617 |
+
"hash_full_prompts": "e34a028d0ddeec5e",
|
| 2618 |
+
"hash_input_tokens": "7c7f37778e6ccda2",
|
| 2619 |
+
"hash_cont_tokens": "f4590c58f12f2766"
|
| 2620 |
+
},
|
| 2621 |
+
"truncated": 0,
|
| 2622 |
+
"non_truncated": 204,
|
| 2623 |
+
"padded": 816,
|
| 2624 |
+
"non_padded": 0,
|
| 2625 |
+
"effective_few_shots": 5.0,
|
| 2626 |
+
"num_truncated_few_shots": 0
|
| 2627 |
+
},
|
| 2628 |
+
"leaderboard|mmlu:high_school_world_history|5": {
|
| 2629 |
+
"hashes": {
|
| 2630 |
+
"hash_examples": "7e5085b6184b0322",
|
| 2631 |
+
"hash_full_prompts": "1fa3d51392765601",
|
| 2632 |
+
"hash_input_tokens": "71993d416140265b",
|
| 2633 |
+
"hash_cont_tokens": "db6bcddd891df5d9"
|
| 2634 |
+
},
|
| 2635 |
+
"truncated": 0,
|
| 2636 |
+
"non_truncated": 237,
|
| 2637 |
+
"padded": 948,
|
| 2638 |
+
"non_padded": 0,
|
| 2639 |
+
"effective_few_shots": 5.0,
|
| 2640 |
+
"num_truncated_few_shots": 0
|
| 2641 |
+
},
|
| 2642 |
+
"leaderboard|mmlu:human_aging|5": {
|
| 2643 |
+
"hashes": {
|
| 2644 |
+
"hash_examples": "c17333e7c7c10797",
|
| 2645 |
+
"hash_full_prompts": "cac900721f9a1a94",
|
| 2646 |
+
"hash_input_tokens": "b0fa52119d4303e9",
|
| 2647 |
+
"hash_cont_tokens": "25cec8d640319105"
|
| 2648 |
+
},
|
| 2649 |
+
"truncated": 0,
|
| 2650 |
+
"non_truncated": 223,
|
| 2651 |
+
"padded": 892,
|
| 2652 |
+
"non_padded": 0,
|
| 2653 |
+
"effective_few_shots": 5.0,
|
| 2654 |
+
"num_truncated_few_shots": 0
|
| 2655 |
+
},
|
| 2656 |
+
"leaderboard|mmlu:human_sexuality|5": {
|
| 2657 |
+
"hashes": {
|
| 2658 |
+
"hash_examples": "4edd1e9045df5e3d",
|
| 2659 |
+
"hash_full_prompts": "0d6567bafee0a13c",
|
| 2660 |
+
"hash_input_tokens": "879018ae27bdf5b0",
|
| 2661 |
+
"hash_cont_tokens": "6778302b4a10b645"
|
| 2662 |
+
},
|
| 2663 |
+
"truncated": 0,
|
| 2664 |
+
"non_truncated": 131,
|
| 2665 |
+
"padded": 524,
|
| 2666 |
+
"non_padded": 0,
|
| 2667 |
+
"effective_few_shots": 5.0,
|
| 2668 |
+
"num_truncated_few_shots": 0
|
| 2669 |
+
},
|
| 2670 |
+
"leaderboard|mmlu:international_law|5": {
|
| 2671 |
+
"hashes": {
|
| 2672 |
+
"hash_examples": "db2fa00d771a062a",
|
| 2673 |
+
"hash_full_prompts": "d018f9116479795e",
|
| 2674 |
+
"hash_input_tokens": "be4409fc3ab936f3",
|
| 2675 |
+
"hash_cont_tokens": "9eb54e1a46032749"
|
| 2676 |
+
},
|
| 2677 |
+
"truncated": 0,
|
| 2678 |
+
"non_truncated": 121,
|
| 2679 |
+
"padded": 484,
|
| 2680 |
+
"non_padded": 0,
|
| 2681 |
+
"effective_few_shots": 5.0,
|
| 2682 |
+
"num_truncated_few_shots": 0
|
| 2683 |
+
},
|
| 2684 |
+
"leaderboard|mmlu:jurisprudence|5": {
|
| 2685 |
+
"hashes": {
|
| 2686 |
+
"hash_examples": "e956f86b124076fe",
|
| 2687 |
+
"hash_full_prompts": "1487e89a10ec58b7",
|
| 2688 |
+
"hash_input_tokens": "888c2eab4655e553",
|
| 2689 |
+
"hash_cont_tokens": "f17d9a372cfd66b1"
|
| 2690 |
+
},
|
| 2691 |
+
"truncated": 0,
|
| 2692 |
+
"non_truncated": 108,
|
| 2693 |
+
"padded": 420,
|
| 2694 |
+
"non_padded": 12,
|
| 2695 |
+
"effective_few_shots": 5.0,
|
| 2696 |
+
"num_truncated_few_shots": 0
|
| 2697 |
+
},
|
| 2698 |
+
"leaderboard|mmlu:logical_fallacies|5": {
|
| 2699 |
+
"hashes": {
|
| 2700 |
+
"hash_examples": "956e0e6365ab79f1",
|
| 2701 |
+
"hash_full_prompts": "677785b2181f9243",
|
| 2702 |
+
"hash_input_tokens": "8cee26c610ab13a1",
|
| 2703 |
+
"hash_cont_tokens": "cf44a68f5bca9a96"
|
| 2704 |
+
},
|
| 2705 |
+
"truncated": 0,
|
| 2706 |
+
"non_truncated": 163,
|
| 2707 |
+
"padded": 648,
|
| 2708 |
+
"non_padded": 4,
|
| 2709 |
+
"effective_few_shots": 5.0,
|
| 2710 |
+
"num_truncated_few_shots": 0
|
| 2711 |
+
},
|
| 2712 |
+
"leaderboard|mmlu:machine_learning|5": {
|
| 2713 |
+
"hashes": {
|
| 2714 |
+
"hash_examples": "397997cc6f4d581e",
|
| 2715 |
+
"hash_full_prompts": "769ee14a2aea49bb",
|
| 2716 |
+
"hash_input_tokens": "1d8a213f41f96aee",
|
| 2717 |
+
"hash_cont_tokens": "eace00d420f4f32c"
|
| 2718 |
+
},
|
| 2719 |
+
"truncated": 0,
|
| 2720 |
+
"non_truncated": 112,
|
| 2721 |
+
"padded": 448,
|
| 2722 |
+
"non_padded": 0,
|
| 2723 |
+
"effective_few_shots": 5.0,
|
| 2724 |
+
"num_truncated_few_shots": 0
|
| 2725 |
+
},
|
| 2726 |
+
"leaderboard|mmlu:management|5": {
|
| 2727 |
+
"hashes": {
|
| 2728 |
+
"hash_examples": "2bcbe6f6ca63d740",
|
| 2729 |
+
"hash_full_prompts": "cb1ff9dac9582144",
|
| 2730 |
+
"hash_input_tokens": "44ba435973dce9d1",
|
| 2731 |
+
"hash_cont_tokens": "b7c51d0250c252d8"
|
| 2732 |
+
},
|
| 2733 |
+
"truncated": 0,
|
| 2734 |
+
"non_truncated": 103,
|
| 2735 |
+
"padded": 412,
|
| 2736 |
+
"non_padded": 0,
|
| 2737 |
+
"effective_few_shots": 5.0,
|
| 2738 |
+
"num_truncated_few_shots": 0
|
| 2739 |
+
},
|
| 2740 |
+
"leaderboard|mmlu:marketing|5": {
|
| 2741 |
+
"hashes": {
|
| 2742 |
+
"hash_examples": "8ddb20d964a1b065",
|
| 2743 |
+
"hash_full_prompts": "9fc2114a187ad9a2",
|
| 2744 |
+
"hash_input_tokens": "e86c7f7e4f27bcb7",
|
| 2745 |
+
"hash_cont_tokens": "086fb63f8b1d1339"
|
| 2746 |
+
},
|
| 2747 |
+
"truncated": 0,
|
| 2748 |
+
"non_truncated": 234,
|
| 2749 |
+
"padded": 924,
|
| 2750 |
+
"non_padded": 12,
|
| 2751 |
+
"effective_few_shots": 5.0,
|
| 2752 |
+
"num_truncated_few_shots": 0
|
| 2753 |
+
},
|
| 2754 |
+
"leaderboard|mmlu:medical_genetics|5": {
|
| 2755 |
+
"hashes": {
|
| 2756 |
+
"hash_examples": "182a71f4763d2cea",
|
| 2757 |
+
"hash_full_prompts": "46a616fa51878959",
|
| 2758 |
+
"hash_input_tokens": "84615035f844ffa0",
|
| 2759 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2760 |
+
},
|
| 2761 |
+
"truncated": 0,
|
| 2762 |
+
"non_truncated": 100,
|
| 2763 |
+
"padded": 400,
|
| 2764 |
+
"non_padded": 0,
|
| 2765 |
+
"effective_few_shots": 5.0,
|
| 2766 |
+
"num_truncated_few_shots": 0
|
| 2767 |
+
},
|
| 2768 |
+
"leaderboard|mmlu:miscellaneous|5": {
|
| 2769 |
+
"hashes": {
|
| 2770 |
+
"hash_examples": "4c404fdbb4ca57fc",
|
| 2771 |
+
"hash_full_prompts": "0813e1be36dbaae1",
|
| 2772 |
+
"hash_input_tokens": "f816152d0e727938",
|
| 2773 |
+
"hash_cont_tokens": "1827274fa6537077"
|
| 2774 |
+
},
|
| 2775 |
+
"truncated": 0,
|
| 2776 |
+
"non_truncated": 783,
|
| 2777 |
+
"padded": 3132,
|
| 2778 |
+
"non_padded": 0,
|
| 2779 |
+
"effective_few_shots": 5.0,
|
| 2780 |
+
"num_truncated_few_shots": 0
|
| 2781 |
+
},
|
| 2782 |
+
"leaderboard|mmlu:moral_disputes|5": {
|
| 2783 |
+
"hashes": {
|
| 2784 |
+
"hash_examples": "60cbd2baa3fea5c9",
|
| 2785 |
+
"hash_full_prompts": "1d14adebb9b62519",
|
| 2786 |
+
"hash_input_tokens": "53082748f1b5e440",
|
| 2787 |
+
"hash_cont_tokens": "472c223f6f28cfc7"
|
| 2788 |
+
},
|
| 2789 |
+
"truncated": 0,
|
| 2790 |
+
"non_truncated": 346,
|
| 2791 |
+
"padded": 1384,
|
| 2792 |
+
"non_padded": 0,
|
| 2793 |
+
"effective_few_shots": 5.0,
|
| 2794 |
+
"num_truncated_few_shots": 0
|
| 2795 |
+
},
|
| 2796 |
+
"leaderboard|mmlu:moral_scenarios|5": {
|
| 2797 |
+
"hashes": {
|
| 2798 |
+
"hash_examples": "fd8b0431fbdd75ef",
|
| 2799 |
+
"hash_full_prompts": "b80d3d236165e3de",
|
| 2800 |
+
"hash_input_tokens": "b5318303d9c36325",
|
| 2801 |
+
"hash_cont_tokens": "e90dade00a092f9e"
|
| 2802 |
+
},
|
| 2803 |
+
"truncated": 0,
|
| 2804 |
+
"non_truncated": 895,
|
| 2805 |
+
"padded": 3567,
|
| 2806 |
+
"non_padded": 13,
|
| 2807 |
+
"effective_few_shots": 5.0,
|
| 2808 |
+
"num_truncated_few_shots": 0
|
| 2809 |
+
},
|
| 2810 |
+
"leaderboard|mmlu:nutrition|5": {
|
| 2811 |
+
"hashes": {
|
| 2812 |
+
"hash_examples": "71e55e2b829b6528",
|
| 2813 |
+
"hash_full_prompts": "2bfb18e5fab8dea7",
|
| 2814 |
+
"hash_input_tokens": "2ed8503c57d6afbf",
|
| 2815 |
+
"hash_cont_tokens": "128e0ec97d96b165"
|
| 2816 |
+
},
|
| 2817 |
+
"truncated": 0,
|
| 2818 |
+
"non_truncated": 306,
|
| 2819 |
+
"padded": 1224,
|
| 2820 |
+
"non_padded": 0,
|
| 2821 |
+
"effective_few_shots": 5.0,
|
| 2822 |
+
"num_truncated_few_shots": 0
|
| 2823 |
+
},
|
| 2824 |
+
"leaderboard|mmlu:philosophy|5": {
|
| 2825 |
+
"hashes": {
|
| 2826 |
+
"hash_examples": "a6d489a8d208fa4b",
|
| 2827 |
+
"hash_full_prompts": "e8c0d5b6dae3ccc8",
|
| 2828 |
+
"hash_input_tokens": "7e8ad59a08a00f3b",
|
| 2829 |
+
"hash_cont_tokens": "cbfd7829a3e0f082"
|
| 2830 |
+
},
|
| 2831 |
+
"truncated": 0,
|
| 2832 |
+
"non_truncated": 311,
|
| 2833 |
+
"padded": 1244,
|
| 2834 |
+
"non_padded": 0,
|
| 2835 |
+
"effective_few_shots": 5.0,
|
| 2836 |
+
"num_truncated_few_shots": 0
|
| 2837 |
+
},
|
| 2838 |
+
"leaderboard|mmlu:prehistory|5": {
|
| 2839 |
+
"hashes": {
|
| 2840 |
+
"hash_examples": "6cc50f032a19acaa",
|
| 2841 |
+
"hash_full_prompts": "4a6a1d3ab1bf28e4",
|
| 2842 |
+
"hash_input_tokens": "8bba5be57a92c467",
|
| 2843 |
+
"hash_cont_tokens": "9c0cf5a2f71afa7e"
|
| 2844 |
+
},
|
| 2845 |
+
"truncated": 0,
|
| 2846 |
+
"non_truncated": 324,
|
| 2847 |
+
"padded": 1284,
|
| 2848 |
+
"non_padded": 12,
|
| 2849 |
+
"effective_few_shots": 5.0,
|
| 2850 |
+
"num_truncated_few_shots": 0
|
| 2851 |
+
},
|
| 2852 |
+
"leaderboard|mmlu:professional_accounting|5": {
|
| 2853 |
+
"hashes": {
|
| 2854 |
+
"hash_examples": "50f57ab32f5f6cea",
|
| 2855 |
+
"hash_full_prompts": "e60129bd2d82ffc6",
|
| 2856 |
+
"hash_input_tokens": "236927cb4e27f724",
|
| 2857 |
+
"hash_cont_tokens": "50f011c2453517ee"
|
| 2858 |
+
},
|
| 2859 |
+
"truncated": 0,
|
| 2860 |
+
"non_truncated": 282,
|
| 2861 |
+
"padded": 1128,
|
| 2862 |
+
"non_padded": 0,
|
| 2863 |
+
"effective_few_shots": 5.0,
|
| 2864 |
+
"num_truncated_few_shots": 0
|
| 2865 |
+
},
|
| 2866 |
+
"leaderboard|mmlu:professional_law|5": {
|
| 2867 |
+
"hashes": {
|
| 2868 |
+
"hash_examples": "a8fdc85c64f4b215",
|
| 2869 |
+
"hash_full_prompts": "0dbb1d9b72dcea03",
|
| 2870 |
+
"hash_input_tokens": "7958ac5eb01fed27",
|
| 2871 |
+
"hash_cont_tokens": "73527e852c24186c"
|
| 2872 |
+
},
|
| 2873 |
+
"truncated": 0,
|
| 2874 |
+
"non_truncated": 1534,
|
| 2875 |
+
"padded": 6136,
|
| 2876 |
+
"non_padded": 0,
|
| 2877 |
+
"effective_few_shots": 5.0,
|
| 2878 |
+
"num_truncated_few_shots": 0
|
| 2879 |
+
},
|
| 2880 |
+
"leaderboard|mmlu:professional_medicine|5": {
|
| 2881 |
+
"hashes": {
|
| 2882 |
+
"hash_examples": "c373a28a3050a73a",
|
| 2883 |
+
"hash_full_prompts": "5e040f9ca68b089e",
|
| 2884 |
+
"hash_input_tokens": "f520600f7896a87b",
|
| 2885 |
+
"hash_cont_tokens": "ceb7af5e2e789abc"
|
| 2886 |
+
},
|
| 2887 |
+
"truncated": 0,
|
| 2888 |
+
"non_truncated": 272,
|
| 2889 |
+
"padded": 1088,
|
| 2890 |
+
"non_padded": 0,
|
| 2891 |
+
"effective_few_shots": 5.0,
|
| 2892 |
+
"num_truncated_few_shots": 0
|
| 2893 |
+
},
|
| 2894 |
+
"leaderboard|mmlu:professional_psychology|5": {
|
| 2895 |
+
"hashes": {
|
| 2896 |
+
"hash_examples": "bf5254fe818356af",
|
| 2897 |
+
"hash_full_prompts": "b386ecda8b87150e",
|
| 2898 |
+
"hash_input_tokens": "fb3f225a047d0f0f",
|
| 2899 |
+
"hash_cont_tokens": "8cfdced8a9667380"
|
| 2900 |
+
},
|
| 2901 |
+
"truncated": 0,
|
| 2902 |
+
"non_truncated": 612,
|
| 2903 |
+
"padded": 2428,
|
| 2904 |
+
"non_padded": 20,
|
| 2905 |
+
"effective_few_shots": 5.0,
|
| 2906 |
+
"num_truncated_few_shots": 0
|
| 2907 |
+
},
|
| 2908 |
+
"leaderboard|mmlu:public_relations|5": {
|
| 2909 |
+
"hashes": {
|
| 2910 |
+
"hash_examples": "b66d52e28e7d14e0",
|
| 2911 |
+
"hash_full_prompts": "fe43562263e25677",
|
| 2912 |
+
"hash_input_tokens": "9dfb929ef5e3362b",
|
| 2913 |
+
"hash_cont_tokens": "f8327461a9cc5123"
|
| 2914 |
+
},
|
| 2915 |
+
"truncated": 0,
|
| 2916 |
+
"non_truncated": 110,
|
| 2917 |
+
"padded": 436,
|
| 2918 |
+
"non_padded": 4,
|
| 2919 |
+
"effective_few_shots": 5.0,
|
| 2920 |
+
"num_truncated_few_shots": 0
|
| 2921 |
+
},
|
| 2922 |
+
"leaderboard|mmlu:security_studies|5": {
|
| 2923 |
+
"hashes": {
|
| 2924 |
+
"hash_examples": "514c14feaf000ad9",
|
| 2925 |
+
"hash_full_prompts": "27d4a2ac541ef4b9",
|
| 2926 |
+
"hash_input_tokens": "f620744b07919b24",
|
| 2927 |
+
"hash_cont_tokens": "c30b0c4d52c2875d"
|
| 2928 |
+
},
|
| 2929 |
+
"truncated": 0,
|
| 2930 |
+
"non_truncated": 245,
|
| 2931 |
+
"padded": 980,
|
| 2932 |
+
"non_padded": 0,
|
| 2933 |
+
"effective_few_shots": 5.0,
|
| 2934 |
+
"num_truncated_few_shots": 0
|
| 2935 |
+
},
|
| 2936 |
+
"leaderboard|mmlu:sociology|5": {
|
| 2937 |
+
"hashes": {
|
| 2938 |
+
"hash_examples": "f6c9bc9d18c80870",
|
| 2939 |
+
"hash_full_prompts": "c072ea7d1a1524f2",
|
| 2940 |
+
"hash_input_tokens": "76d03f98f30dbe11",
|
| 2941 |
+
"hash_cont_tokens": "eef4bd16d536fbd6"
|
| 2942 |
+
},
|
| 2943 |
+
"truncated": 0,
|
| 2944 |
+
"non_truncated": 201,
|
| 2945 |
+
"padded": 804,
|
| 2946 |
+
"non_padded": 0,
|
| 2947 |
+
"effective_few_shots": 5.0,
|
| 2948 |
+
"num_truncated_few_shots": 0
|
| 2949 |
+
},
|
| 2950 |
+
"leaderboard|mmlu:us_foreign_policy|5": {
|
| 2951 |
+
"hashes": {
|
| 2952 |
+
"hash_examples": "ed7b78629db6678f",
|
| 2953 |
+
"hash_full_prompts": "341a97ca3e4d699d",
|
| 2954 |
+
"hash_input_tokens": "f0b4b93f91f3d7f4",
|
| 2955 |
+
"hash_cont_tokens": "00520b0ec06da34f"
|
| 2956 |
+
},
|
| 2957 |
+
"truncated": 0,
|
| 2958 |
+
"non_truncated": 100,
|
| 2959 |
+
"padded": 400,
|
| 2960 |
+
"non_padded": 0,
|
| 2961 |
+
"effective_few_shots": 5.0,
|
| 2962 |
+
"num_truncated_few_shots": 0
|
| 2963 |
+
},
|
| 2964 |
+
"leaderboard|mmlu:virology|5": {
|
| 2965 |
+
"hashes": {
|
| 2966 |
+
"hash_examples": "bc52ffdc3f9b994a",
|
| 2967 |
+
"hash_full_prompts": "651d471e2eb8b5e9",
|
| 2968 |
+
"hash_input_tokens": "1c7d23a204c7cbf6",
|
| 2969 |
+
"hash_cont_tokens": "f5fc195e049353c0"
|
| 2970 |
+
},
|
| 2971 |
+
"truncated": 0,
|
| 2972 |
+
"non_truncated": 166,
|
| 2973 |
+
"padded": 664,
|
| 2974 |
+
"non_padded": 0,
|
| 2975 |
+
"effective_few_shots": 5.0,
|
| 2976 |
+
"num_truncated_few_shots": 0
|
| 2977 |
+
},
|
| 2978 |
+
"leaderboard|mmlu:world_religions|5": {
|
| 2979 |
+
"hashes": {
|
| 2980 |
+
"hash_examples": "ecdb4a4f94f62930",
|
| 2981 |
+
"hash_full_prompts": "3773f03542ce44a3",
|
| 2982 |
+
"hash_input_tokens": "be42fd2c9cc2da08",
|
| 2983 |
+
"hash_cont_tokens": "ada548665e87b1e0"
|
| 2984 |
+
},
|
| 2985 |
+
"truncated": 0,
|
| 2986 |
+
"non_truncated": 171,
|
| 2987 |
+
"padded": 684,
|
| 2988 |
+
"non_padded": 0,
|
| 2989 |
+
"effective_few_shots": 5.0,
|
| 2990 |
+
"num_truncated_few_shots": 0
|
| 2991 |
+
}
|
| 2992 |
+
},
|
| 2993 |
+
"summary_general": {
|
| 2994 |
+
"hashes": {
|
| 2995 |
+
"hash_examples": "341a076d0beb7048",
|
| 2996 |
+
"hash_full_prompts": "a5c8f2b7ff4f5ae2",
|
| 2997 |
+
"hash_input_tokens": "917c40aba1546e12",
|
| 2998 |
+
"hash_cont_tokens": "3672212ca582e2d0"
|
| 2999 |
+
},
|
| 3000 |
+
"truncated": 0,
|
| 3001 |
+
"non_truncated": 14042,
|
| 3002 |
+
"padded": 56038,
|
| 3003 |
+
"non_padded": 130,
|
| 3004 |
+
"num_truncated_few_shots": 0
|
| 3005 |
+
}
|
| 3006 |
+
}
|