Upload eval_results/HuggingFaceH4/mistral-7b-ift/v31.3/eval_mmlu.json with huggingface_hub
Browse files
eval_results/HuggingFaceH4/mistral-7b-ift/v31.3/eval_mmlu.json
ADDED
|
@@ -0,0 +1,2651 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"mmlu": {
|
| 4 |
+
"acc,none": 0.6290414470873095,
|
| 5 |
+
"acc_stderr,none": 0.0038616186138953307,
|
| 6 |
+
"alias": "mmlu"
|
| 7 |
+
},
|
| 8 |
+
"mmlu_humanities": {
|
| 9 |
+
"alias": " - humanities",
|
| 10 |
+
"acc,none": 0.5827842720510096,
|
| 11 |
+
"acc_stderr,none": 0.006787513340184287
|
| 12 |
+
},
|
| 13 |
+
"mmlu_formal_logic": {
|
| 14 |
+
"alias": " - formal_logic",
|
| 15 |
+
"acc,none": 0.4444444444444444,
|
| 16 |
+
"acc_stderr,none": 0.044444444444444495
|
| 17 |
+
},
|
| 18 |
+
"mmlu_high_school_european_history": {
|
| 19 |
+
"alias": " - high_school_european_history",
|
| 20 |
+
"acc,none": 0.7696969696969697,
|
| 21 |
+
"acc_stderr,none": 0.03287666758603488
|
| 22 |
+
},
|
| 23 |
+
"mmlu_high_school_us_history": {
|
| 24 |
+
"alias": " - high_school_us_history",
|
| 25 |
+
"acc,none": 0.8186274509803921,
|
| 26 |
+
"acc_stderr,none": 0.027044621719474082
|
| 27 |
+
},
|
| 28 |
+
"mmlu_high_school_world_history": {
|
| 29 |
+
"alias": " - high_school_world_history",
|
| 30 |
+
"acc,none": 0.7974683544303798,
|
| 31 |
+
"acc_stderr,none": 0.026160568246601446
|
| 32 |
+
},
|
| 33 |
+
"mmlu_international_law": {
|
| 34 |
+
"alias": " - international_law",
|
| 35 |
+
"acc,none": 0.768595041322314,
|
| 36 |
+
"acc_stderr,none": 0.03849856098794087
|
| 37 |
+
},
|
| 38 |
+
"mmlu_jurisprudence": {
|
| 39 |
+
"alias": " - jurisprudence",
|
| 40 |
+
"acc,none": 0.7592592592592593,
|
| 41 |
+
"acc_stderr,none": 0.04133119440243838
|
| 42 |
+
},
|
| 43 |
+
"mmlu_logical_fallacies": {
|
| 44 |
+
"alias": " - logical_fallacies",
|
| 45 |
+
"acc,none": 0.7730061349693251,
|
| 46 |
+
"acc_stderr,none": 0.03291099578615769
|
| 47 |
+
},
|
| 48 |
+
"mmlu_moral_disputes": {
|
| 49 |
+
"alias": " - moral_disputes",
|
| 50 |
+
"acc,none": 0.7225433526011561,
|
| 51 |
+
"acc_stderr,none": 0.024105712607754307
|
| 52 |
+
},
|
| 53 |
+
"mmlu_moral_scenarios": {
|
| 54 |
+
"alias": " - moral_scenarios",
|
| 55 |
+
"acc,none": 0.39776536312849164,
|
| 56 |
+
"acc_stderr,none": 0.016369204971262985
|
| 57 |
+
},
|
| 58 |
+
"mmlu_philosophy": {
|
| 59 |
+
"alias": " - philosophy",
|
| 60 |
+
"acc,none": 0.7202572347266881,
|
| 61 |
+
"acc_stderr,none": 0.025494259350694912
|
| 62 |
+
},
|
| 63 |
+
"mmlu_prehistory": {
|
| 64 |
+
"alias": " - prehistory",
|
| 65 |
+
"acc,none": 0.7160493827160493,
|
| 66 |
+
"acc_stderr,none": 0.02508947852376513
|
| 67 |
+
},
|
| 68 |
+
"mmlu_professional_law": {
|
| 69 |
+
"alias": " - professional_law",
|
| 70 |
+
"acc,none": 0.45632333767926986,
|
| 71 |
+
"acc_stderr,none": 0.012721420501462547
|
| 72 |
+
},
|
| 73 |
+
"mmlu_world_religions": {
|
| 74 |
+
"alias": " - world_religions",
|
| 75 |
+
"acc,none": 0.8187134502923976,
|
| 76 |
+
"acc_stderr,none": 0.029547741687640038
|
| 77 |
+
},
|
| 78 |
+
"mmlu_other": {
|
| 79 |
+
"alias": " - other",
|
| 80 |
+
"acc,none": 0.6961699388477631,
|
| 81 |
+
"acc_stderr,none": 0.00794157385703638
|
| 82 |
+
},
|
| 83 |
+
"mmlu_business_ethics": {
|
| 84 |
+
"alias": " - business_ethics",
|
| 85 |
+
"acc,none": 0.59,
|
| 86 |
+
"acc_stderr,none": 0.04943110704237102
|
| 87 |
+
},
|
| 88 |
+
"mmlu_clinical_knowledge": {
|
| 89 |
+
"alias": " - clinical_knowledge",
|
| 90 |
+
"acc,none": 0.6981132075471698,
|
| 91 |
+
"acc_stderr,none": 0.02825420034443866
|
| 92 |
+
},
|
| 93 |
+
"mmlu_college_medicine": {
|
| 94 |
+
"alias": " - college_medicine",
|
| 95 |
+
"acc,none": 0.653179190751445,
|
| 96 |
+
"acc_stderr,none": 0.03629146670159665
|
| 97 |
+
},
|
| 98 |
+
"mmlu_global_facts": {
|
| 99 |
+
"alias": " - global_facts",
|
| 100 |
+
"acc,none": 0.39,
|
| 101 |
+
"acc_stderr,none": 0.04902071300001975
|
| 102 |
+
},
|
| 103 |
+
"mmlu_human_aging": {
|
| 104 |
+
"alias": " - human_aging",
|
| 105 |
+
"acc,none": 0.6636771300448431,
|
| 106 |
+
"acc_stderr,none": 0.031708824268455005
|
| 107 |
+
},
|
| 108 |
+
"mmlu_management": {
|
| 109 |
+
"alias": " - management",
|
| 110 |
+
"acc,none": 0.8155339805825242,
|
| 111 |
+
"acc_stderr,none": 0.03840423627288276
|
| 112 |
+
},
|
| 113 |
+
"mmlu_marketing": {
|
| 114 |
+
"alias": " - marketing",
|
| 115 |
+
"acc,none": 0.8760683760683761,
|
| 116 |
+
"acc_stderr,none": 0.021586494001281348
|
| 117 |
+
},
|
| 118 |
+
"mmlu_medical_genetics": {
|
| 119 |
+
"alias": " - medical_genetics",
|
| 120 |
+
"acc,none": 0.74,
|
| 121 |
+
"acc_stderr,none": 0.04408440022768079
|
| 122 |
+
},
|
| 123 |
+
"mmlu_miscellaneous": {
|
| 124 |
+
"alias": " - miscellaneous",
|
| 125 |
+
"acc,none": 0.8020434227330779,
|
| 126 |
+
"acc_stderr,none": 0.01424887354921756
|
| 127 |
+
},
|
| 128 |
+
"mmlu_nutrition": {
|
| 129 |
+
"alias": " - nutrition",
|
| 130 |
+
"acc,none": 0.7483660130718954,
|
| 131 |
+
"acc_stderr,none": 0.0248480182638752
|
| 132 |
+
},
|
| 133 |
+
"mmlu_professional_accounting": {
|
| 134 |
+
"alias": " - professional_accounting",
|
| 135 |
+
"acc,none": 0.45390070921985815,
|
| 136 |
+
"acc_stderr,none": 0.029700453247291463
|
| 137 |
+
},
|
| 138 |
+
"mmlu_professional_medicine": {
|
| 139 |
+
"alias": " - professional_medicine",
|
| 140 |
+
"acc,none": 0.6691176470588235,
|
| 141 |
+
"acc_stderr,none": 0.028582709753898435
|
| 142 |
+
},
|
| 143 |
+
"mmlu_virology": {
|
| 144 |
+
"alias": " - virology",
|
| 145 |
+
"acc,none": 0.536144578313253,
|
| 146 |
+
"acc_stderr,none": 0.03882310850890593
|
| 147 |
+
},
|
| 148 |
+
"mmlu_social_sciences": {
|
| 149 |
+
"alias": " - social_sciences",
|
| 150 |
+
"acc,none": 0.7312317192070198,
|
| 151 |
+
"acc_stderr,none": 0.007850188394169316
|
| 152 |
+
},
|
| 153 |
+
"mmlu_econometrics": {
|
| 154 |
+
"alias": " - econometrics",
|
| 155 |
+
"acc,none": 0.5175438596491229,
|
| 156 |
+
"acc_stderr,none": 0.04700708033551038
|
| 157 |
+
},
|
| 158 |
+
"mmlu_high_school_geography": {
|
| 159 |
+
"alias": " - high_school_geography",
|
| 160 |
+
"acc,none": 0.7777777777777778,
|
| 161 |
+
"acc_stderr,none": 0.029620227874790486
|
| 162 |
+
},
|
| 163 |
+
"mmlu_high_school_government_and_politics": {
|
| 164 |
+
"alias": " - high_school_government_and_politics",
|
| 165 |
+
"acc,none": 0.8652849740932642,
|
| 166 |
+
"acc_stderr,none": 0.024639789097709443
|
| 167 |
+
},
|
| 168 |
+
"mmlu_high_school_macroeconomics": {
|
| 169 |
+
"alias": " - high_school_macroeconomics",
|
| 170 |
+
"acc,none": 0.6564102564102564,
|
| 171 |
+
"acc_stderr,none": 0.024078696580635474
|
| 172 |
+
},
|
| 173 |
+
"mmlu_high_school_microeconomics": {
|
| 174 |
+
"alias": " - high_school_microeconomics",
|
| 175 |
+
"acc,none": 0.6722689075630253,
|
| 176 |
+
"acc_stderr,none": 0.030489911417673227
|
| 177 |
+
},
|
| 178 |
+
"mmlu_high_school_psychology": {
|
| 179 |
+
"alias": " - high_school_psychology",
|
| 180 |
+
"acc,none": 0.8275229357798165,
|
| 181 |
+
"acc_stderr,none": 0.01619780795684803
|
| 182 |
+
},
|
| 183 |
+
"mmlu_human_sexuality": {
|
| 184 |
+
"alias": " - human_sexuality",
|
| 185 |
+
"acc,none": 0.7709923664122137,
|
| 186 |
+
"acc_stderr,none": 0.036853466317118506
|
| 187 |
+
},
|
| 188 |
+
"mmlu_professional_psychology": {
|
| 189 |
+
"alias": " - professional_psychology",
|
| 190 |
+
"acc,none": 0.6683006535947712,
|
| 191 |
+
"acc_stderr,none": 0.019047485239360385
|
| 192 |
+
},
|
| 193 |
+
"mmlu_public_relations": {
|
| 194 |
+
"alias": " - public_relations",
|
| 195 |
+
"acc,none": 0.6272727272727273,
|
| 196 |
+
"acc_stderr,none": 0.04631381319425463
|
| 197 |
+
},
|
| 198 |
+
"mmlu_security_studies": {
|
| 199 |
+
"alias": " - security_studies",
|
| 200 |
+
"acc,none": 0.7224489795918367,
|
| 201 |
+
"acc_stderr,none": 0.028666857790274648
|
| 202 |
+
},
|
| 203 |
+
"mmlu_sociology": {
|
| 204 |
+
"alias": " - sociology",
|
| 205 |
+
"acc,none": 0.8059701492537313,
|
| 206 |
+
"acc_stderr,none": 0.027962677604768914
|
| 207 |
+
},
|
| 208 |
+
"mmlu_us_foreign_policy": {
|
| 209 |
+
"alias": " - us_foreign_policy",
|
| 210 |
+
"acc,none": 0.85,
|
| 211 |
+
"acc_stderr,none": 0.03588702812826371
|
| 212 |
+
},
|
| 213 |
+
"mmlu_stem": {
|
| 214 |
+
"alias": " - stem",
|
| 215 |
+
"acc,none": 0.5321915635902316,
|
| 216 |
+
"acc_stderr,none": 0.008558424855038367
|
| 217 |
+
},
|
| 218 |
+
"mmlu_abstract_algebra": {
|
| 219 |
+
"alias": " - abstract_algebra",
|
| 220 |
+
"acc,none": 0.26,
|
| 221 |
+
"acc_stderr,none": 0.044084400227680794
|
| 222 |
+
},
|
| 223 |
+
"mmlu_anatomy": {
|
| 224 |
+
"alias": " - anatomy",
|
| 225 |
+
"acc,none": 0.6370370370370371,
|
| 226 |
+
"acc_stderr,none": 0.04153948404742398
|
| 227 |
+
},
|
| 228 |
+
"mmlu_astronomy": {
|
| 229 |
+
"alias": " - astronomy",
|
| 230 |
+
"acc,none": 0.6842105263157895,
|
| 231 |
+
"acc_stderr,none": 0.0378272898086547
|
| 232 |
+
},
|
| 233 |
+
"mmlu_college_biology": {
|
| 234 |
+
"alias": " - college_biology",
|
| 235 |
+
"acc,none": 0.7222222222222222,
|
| 236 |
+
"acc_stderr,none": 0.03745554791462457
|
| 237 |
+
},
|
| 238 |
+
"mmlu_college_chemistry": {
|
| 239 |
+
"alias": " - college_chemistry",
|
| 240 |
+
"acc,none": 0.52,
|
| 241 |
+
"acc_stderr,none": 0.050211673156867795
|
| 242 |
+
},
|
| 243 |
+
"mmlu_college_computer_science": {
|
| 244 |
+
"alias": " - college_computer_science",
|
| 245 |
+
"acc,none": 0.53,
|
| 246 |
+
"acc_stderr,none": 0.050161355804659205
|
| 247 |
+
},
|
| 248 |
+
"mmlu_college_mathematics": {
|
| 249 |
+
"alias": " - college_mathematics",
|
| 250 |
+
"acc,none": 0.38,
|
| 251 |
+
"acc_stderr,none": 0.048783173121456316
|
| 252 |
+
},
|
| 253 |
+
"mmlu_college_physics": {
|
| 254 |
+
"alias": " - college_physics",
|
| 255 |
+
"acc,none": 0.4411764705882353,
|
| 256 |
+
"acc_stderr,none": 0.04940635630605659
|
| 257 |
+
},
|
| 258 |
+
"mmlu_computer_security": {
|
| 259 |
+
"alias": " - computer_security",
|
| 260 |
+
"acc,none": 0.72,
|
| 261 |
+
"acc_stderr,none": 0.045126085985421276
|
| 262 |
+
},
|
| 263 |
+
"mmlu_conceptual_physics": {
|
| 264 |
+
"alias": " - conceptual_physics",
|
| 265 |
+
"acc,none": 0.5872340425531914,
|
| 266 |
+
"acc_stderr,none": 0.03218471141400351
|
| 267 |
+
},
|
| 268 |
+
"mmlu_electrical_engineering": {
|
| 269 |
+
"alias": " - electrical_engineering",
|
| 270 |
+
"acc,none": 0.5448275862068965,
|
| 271 |
+
"acc_stderr,none": 0.04149886942192117
|
| 272 |
+
},
|
| 273 |
+
"mmlu_elementary_mathematics": {
|
| 274 |
+
"alias": " - elementary_mathematics",
|
| 275 |
+
"acc,none": 0.4126984126984127,
|
| 276 |
+
"acc_stderr,none": 0.025355741263055266
|
| 277 |
+
},
|
| 278 |
+
"mmlu_high_school_biology": {
|
| 279 |
+
"alias": " - high_school_biology",
|
| 280 |
+
"acc,none": 0.7645161290322581,
|
| 281 |
+
"acc_stderr,none": 0.024137632429337714
|
| 282 |
+
},
|
| 283 |
+
"mmlu_high_school_chemistry": {
|
| 284 |
+
"alias": " - high_school_chemistry",
|
| 285 |
+
"acc,none": 0.5270935960591133,
|
| 286 |
+
"acc_stderr,none": 0.03512819077876106
|
| 287 |
+
},
|
| 288 |
+
"mmlu_high_school_computer_science": {
|
| 289 |
+
"alias": " - high_school_computer_science",
|
| 290 |
+
"acc,none": 0.65,
|
| 291 |
+
"acc_stderr,none": 0.047937248544110196
|
| 292 |
+
},
|
| 293 |
+
"mmlu_high_school_mathematics": {
|
| 294 |
+
"alias": " - high_school_mathematics",
|
| 295 |
+
"acc,none": 0.37777777777777777,
|
| 296 |
+
"acc_stderr,none": 0.029560707392465718
|
| 297 |
+
},
|
| 298 |
+
"mmlu_high_school_physics": {
|
| 299 |
+
"alias": " - high_school_physics",
|
| 300 |
+
"acc,none": 0.3443708609271523,
|
| 301 |
+
"acc_stderr,none": 0.038796870240733264
|
| 302 |
+
},
|
| 303 |
+
"mmlu_high_school_statistics": {
|
| 304 |
+
"alias": " - high_school_statistics",
|
| 305 |
+
"acc,none": 0.5185185185185185,
|
| 306 |
+
"acc_stderr,none": 0.03407632093854051
|
| 307 |
+
},
|
| 308 |
+
"mmlu_machine_learning": {
|
| 309 |
+
"alias": " - machine_learning",
|
| 310 |
+
"acc,none": 0.44642857142857145,
|
| 311 |
+
"acc_stderr,none": 0.047184714852195886
|
| 312 |
+
}
|
| 313 |
+
},
|
| 314 |
+
"groups": {
|
| 315 |
+
"mmlu": {
|
| 316 |
+
"acc,none": 0.6290414470873095,
|
| 317 |
+
"acc_stderr,none": 0.0038616186138953307,
|
| 318 |
+
"alias": "mmlu"
|
| 319 |
+
},
|
| 320 |
+
"mmlu_humanities": {
|
| 321 |
+
"alias": " - humanities",
|
| 322 |
+
"acc,none": 0.5827842720510096,
|
| 323 |
+
"acc_stderr,none": 0.006787513340184287
|
| 324 |
+
},
|
| 325 |
+
"mmlu_other": {
|
| 326 |
+
"alias": " - other",
|
| 327 |
+
"acc,none": 0.6961699388477631,
|
| 328 |
+
"acc_stderr,none": 0.00794157385703638
|
| 329 |
+
},
|
| 330 |
+
"mmlu_social_sciences": {
|
| 331 |
+
"alias": " - social_sciences",
|
| 332 |
+
"acc,none": 0.7312317192070198,
|
| 333 |
+
"acc_stderr,none": 0.007850188394169316
|
| 334 |
+
},
|
| 335 |
+
"mmlu_stem": {
|
| 336 |
+
"alias": " - stem",
|
| 337 |
+
"acc,none": 0.5321915635902316,
|
| 338 |
+
"acc_stderr,none": 0.008558424855038367
|
| 339 |
+
}
|
| 340 |
+
},
|
| 341 |
+
"configs": {
|
| 342 |
+
"mmlu_abstract_algebra": {
|
| 343 |
+
"task": "mmlu_abstract_algebra",
|
| 344 |
+
"task_alias": "abstract_algebra",
|
| 345 |
+
"group": "mmlu_stem",
|
| 346 |
+
"group_alias": "stem",
|
| 347 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 348 |
+
"dataset_name": "abstract_algebra",
|
| 349 |
+
"test_split": "test",
|
| 350 |
+
"fewshot_split": "dev",
|
| 351 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 352 |
+
"doc_to_target": "answer",
|
| 353 |
+
"doc_to_choice": [
|
| 354 |
+
"A",
|
| 355 |
+
"B",
|
| 356 |
+
"C",
|
| 357 |
+
"D"
|
| 358 |
+
],
|
| 359 |
+
"description": "The following are multiple choice questions (with answers) about abstract algebra.\n\n",
|
| 360 |
+
"target_delimiter": " ",
|
| 361 |
+
"fewshot_delimiter": "\n\n",
|
| 362 |
+
"fewshot_config": {
|
| 363 |
+
"sampler": "first_n"
|
| 364 |
+
},
|
| 365 |
+
"num_fewshot": 5,
|
| 366 |
+
"metric_list": [
|
| 367 |
+
{
|
| 368 |
+
"metric": "acc",
|
| 369 |
+
"aggregation": "mean",
|
| 370 |
+
"higher_is_better": true
|
| 371 |
+
}
|
| 372 |
+
],
|
| 373 |
+
"output_type": "multiple_choice",
|
| 374 |
+
"repeats": 1,
|
| 375 |
+
"should_decontaminate": false,
|
| 376 |
+
"metadata": {
|
| 377 |
+
"version": 0.0
|
| 378 |
+
}
|
| 379 |
+
},
|
| 380 |
+
"mmlu_anatomy": {
|
| 381 |
+
"task": "mmlu_anatomy",
|
| 382 |
+
"task_alias": "anatomy",
|
| 383 |
+
"group": "mmlu_stem",
|
| 384 |
+
"group_alias": "stem",
|
| 385 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 386 |
+
"dataset_name": "anatomy",
|
| 387 |
+
"test_split": "test",
|
| 388 |
+
"fewshot_split": "dev",
|
| 389 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 390 |
+
"doc_to_target": "answer",
|
| 391 |
+
"doc_to_choice": [
|
| 392 |
+
"A",
|
| 393 |
+
"B",
|
| 394 |
+
"C",
|
| 395 |
+
"D"
|
| 396 |
+
],
|
| 397 |
+
"description": "The following are multiple choice questions (with answers) about anatomy.\n\n",
|
| 398 |
+
"target_delimiter": " ",
|
| 399 |
+
"fewshot_delimiter": "\n\n",
|
| 400 |
+
"fewshot_config": {
|
| 401 |
+
"sampler": "first_n"
|
| 402 |
+
},
|
| 403 |
+
"num_fewshot": 5,
|
| 404 |
+
"metric_list": [
|
| 405 |
+
{
|
| 406 |
+
"metric": "acc",
|
| 407 |
+
"aggregation": "mean",
|
| 408 |
+
"higher_is_better": true
|
| 409 |
+
}
|
| 410 |
+
],
|
| 411 |
+
"output_type": "multiple_choice",
|
| 412 |
+
"repeats": 1,
|
| 413 |
+
"should_decontaminate": false,
|
| 414 |
+
"metadata": {
|
| 415 |
+
"version": 0.0
|
| 416 |
+
}
|
| 417 |
+
},
|
| 418 |
+
"mmlu_astronomy": {
|
| 419 |
+
"task": "mmlu_astronomy",
|
| 420 |
+
"task_alias": "astronomy",
|
| 421 |
+
"group": "mmlu_stem",
|
| 422 |
+
"group_alias": "stem",
|
| 423 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 424 |
+
"dataset_name": "astronomy",
|
| 425 |
+
"test_split": "test",
|
| 426 |
+
"fewshot_split": "dev",
|
| 427 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 428 |
+
"doc_to_target": "answer",
|
| 429 |
+
"doc_to_choice": [
|
| 430 |
+
"A",
|
| 431 |
+
"B",
|
| 432 |
+
"C",
|
| 433 |
+
"D"
|
| 434 |
+
],
|
| 435 |
+
"description": "The following are multiple choice questions (with answers) about astronomy.\n\n",
|
| 436 |
+
"target_delimiter": " ",
|
| 437 |
+
"fewshot_delimiter": "\n\n",
|
| 438 |
+
"fewshot_config": {
|
| 439 |
+
"sampler": "first_n"
|
| 440 |
+
},
|
| 441 |
+
"num_fewshot": 5,
|
| 442 |
+
"metric_list": [
|
| 443 |
+
{
|
| 444 |
+
"metric": "acc",
|
| 445 |
+
"aggregation": "mean",
|
| 446 |
+
"higher_is_better": true
|
| 447 |
+
}
|
| 448 |
+
],
|
| 449 |
+
"output_type": "multiple_choice",
|
| 450 |
+
"repeats": 1,
|
| 451 |
+
"should_decontaminate": false,
|
| 452 |
+
"metadata": {
|
| 453 |
+
"version": 0.0
|
| 454 |
+
}
|
| 455 |
+
},
|
| 456 |
+
"mmlu_business_ethics": {
|
| 457 |
+
"task": "mmlu_business_ethics",
|
| 458 |
+
"task_alias": "business_ethics",
|
| 459 |
+
"group": "mmlu_other",
|
| 460 |
+
"group_alias": "other",
|
| 461 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 462 |
+
"dataset_name": "business_ethics",
|
| 463 |
+
"test_split": "test",
|
| 464 |
+
"fewshot_split": "dev",
|
| 465 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 466 |
+
"doc_to_target": "answer",
|
| 467 |
+
"doc_to_choice": [
|
| 468 |
+
"A",
|
| 469 |
+
"B",
|
| 470 |
+
"C",
|
| 471 |
+
"D"
|
| 472 |
+
],
|
| 473 |
+
"description": "The following are multiple choice questions (with answers) about business ethics.\n\n",
|
| 474 |
+
"target_delimiter": " ",
|
| 475 |
+
"fewshot_delimiter": "\n\n",
|
| 476 |
+
"fewshot_config": {
|
| 477 |
+
"sampler": "first_n"
|
| 478 |
+
},
|
| 479 |
+
"num_fewshot": 5,
|
| 480 |
+
"metric_list": [
|
| 481 |
+
{
|
| 482 |
+
"metric": "acc",
|
| 483 |
+
"aggregation": "mean",
|
| 484 |
+
"higher_is_better": true
|
| 485 |
+
}
|
| 486 |
+
],
|
| 487 |
+
"output_type": "multiple_choice",
|
| 488 |
+
"repeats": 1,
|
| 489 |
+
"should_decontaminate": false,
|
| 490 |
+
"metadata": {
|
| 491 |
+
"version": 0.0
|
| 492 |
+
}
|
| 493 |
+
},
|
| 494 |
+
"mmlu_clinical_knowledge": {
|
| 495 |
+
"task": "mmlu_clinical_knowledge",
|
| 496 |
+
"task_alias": "clinical_knowledge",
|
| 497 |
+
"group": "mmlu_other",
|
| 498 |
+
"group_alias": "other",
|
| 499 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 500 |
+
"dataset_name": "clinical_knowledge",
|
| 501 |
+
"test_split": "test",
|
| 502 |
+
"fewshot_split": "dev",
|
| 503 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 504 |
+
"doc_to_target": "answer",
|
| 505 |
+
"doc_to_choice": [
|
| 506 |
+
"A",
|
| 507 |
+
"B",
|
| 508 |
+
"C",
|
| 509 |
+
"D"
|
| 510 |
+
],
|
| 511 |
+
"description": "The following are multiple choice questions (with answers) about clinical knowledge.\n\n",
|
| 512 |
+
"target_delimiter": " ",
|
| 513 |
+
"fewshot_delimiter": "\n\n",
|
| 514 |
+
"fewshot_config": {
|
| 515 |
+
"sampler": "first_n"
|
| 516 |
+
},
|
| 517 |
+
"num_fewshot": 5,
|
| 518 |
+
"metric_list": [
|
| 519 |
+
{
|
| 520 |
+
"metric": "acc",
|
| 521 |
+
"aggregation": "mean",
|
| 522 |
+
"higher_is_better": true
|
| 523 |
+
}
|
| 524 |
+
],
|
| 525 |
+
"output_type": "multiple_choice",
|
| 526 |
+
"repeats": 1,
|
| 527 |
+
"should_decontaminate": false,
|
| 528 |
+
"metadata": {
|
| 529 |
+
"version": 0.0
|
| 530 |
+
}
|
| 531 |
+
},
|
| 532 |
+
"mmlu_college_biology": {
|
| 533 |
+
"task": "mmlu_college_biology",
|
| 534 |
+
"task_alias": "college_biology",
|
| 535 |
+
"group": "mmlu_stem",
|
| 536 |
+
"group_alias": "stem",
|
| 537 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 538 |
+
"dataset_name": "college_biology",
|
| 539 |
+
"test_split": "test",
|
| 540 |
+
"fewshot_split": "dev",
|
| 541 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 542 |
+
"doc_to_target": "answer",
|
| 543 |
+
"doc_to_choice": [
|
| 544 |
+
"A",
|
| 545 |
+
"B",
|
| 546 |
+
"C",
|
| 547 |
+
"D"
|
| 548 |
+
],
|
| 549 |
+
"description": "The following are multiple choice questions (with answers) about college biology.\n\n",
|
| 550 |
+
"target_delimiter": " ",
|
| 551 |
+
"fewshot_delimiter": "\n\n",
|
| 552 |
+
"fewshot_config": {
|
| 553 |
+
"sampler": "first_n"
|
| 554 |
+
},
|
| 555 |
+
"num_fewshot": 5,
|
| 556 |
+
"metric_list": [
|
| 557 |
+
{
|
| 558 |
+
"metric": "acc",
|
| 559 |
+
"aggregation": "mean",
|
| 560 |
+
"higher_is_better": true
|
| 561 |
+
}
|
| 562 |
+
],
|
| 563 |
+
"output_type": "multiple_choice",
|
| 564 |
+
"repeats": 1,
|
| 565 |
+
"should_decontaminate": false,
|
| 566 |
+
"metadata": {
|
| 567 |
+
"version": 0.0
|
| 568 |
+
}
|
| 569 |
+
},
|
| 570 |
+
"mmlu_college_chemistry": {
|
| 571 |
+
"task": "mmlu_college_chemistry",
|
| 572 |
+
"task_alias": "college_chemistry",
|
| 573 |
+
"group": "mmlu_stem",
|
| 574 |
+
"group_alias": "stem",
|
| 575 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 576 |
+
"dataset_name": "college_chemistry",
|
| 577 |
+
"test_split": "test",
|
| 578 |
+
"fewshot_split": "dev",
|
| 579 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 580 |
+
"doc_to_target": "answer",
|
| 581 |
+
"doc_to_choice": [
|
| 582 |
+
"A",
|
| 583 |
+
"B",
|
| 584 |
+
"C",
|
| 585 |
+
"D"
|
| 586 |
+
],
|
| 587 |
+
"description": "The following are multiple choice questions (with answers) about college chemistry.\n\n",
|
| 588 |
+
"target_delimiter": " ",
|
| 589 |
+
"fewshot_delimiter": "\n\n",
|
| 590 |
+
"fewshot_config": {
|
| 591 |
+
"sampler": "first_n"
|
| 592 |
+
},
|
| 593 |
+
"num_fewshot": 5,
|
| 594 |
+
"metric_list": [
|
| 595 |
+
{
|
| 596 |
+
"metric": "acc",
|
| 597 |
+
"aggregation": "mean",
|
| 598 |
+
"higher_is_better": true
|
| 599 |
+
}
|
| 600 |
+
],
|
| 601 |
+
"output_type": "multiple_choice",
|
| 602 |
+
"repeats": 1,
|
| 603 |
+
"should_decontaminate": false,
|
| 604 |
+
"metadata": {
|
| 605 |
+
"version": 0.0
|
| 606 |
+
}
|
| 607 |
+
},
|
| 608 |
+
"mmlu_college_computer_science": {
|
| 609 |
+
"task": "mmlu_college_computer_science",
|
| 610 |
+
"task_alias": "college_computer_science",
|
| 611 |
+
"group": "mmlu_stem",
|
| 612 |
+
"group_alias": "stem",
|
| 613 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 614 |
+
"dataset_name": "college_computer_science",
|
| 615 |
+
"test_split": "test",
|
| 616 |
+
"fewshot_split": "dev",
|
| 617 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 618 |
+
"doc_to_target": "answer",
|
| 619 |
+
"doc_to_choice": [
|
| 620 |
+
"A",
|
| 621 |
+
"B",
|
| 622 |
+
"C",
|
| 623 |
+
"D"
|
| 624 |
+
],
|
| 625 |
+
"description": "The following are multiple choice questions (with answers) about college computer science.\n\n",
|
| 626 |
+
"target_delimiter": " ",
|
| 627 |
+
"fewshot_delimiter": "\n\n",
|
| 628 |
+
"fewshot_config": {
|
| 629 |
+
"sampler": "first_n"
|
| 630 |
+
},
|
| 631 |
+
"num_fewshot": 5,
|
| 632 |
+
"metric_list": [
|
| 633 |
+
{
|
| 634 |
+
"metric": "acc",
|
| 635 |
+
"aggregation": "mean",
|
| 636 |
+
"higher_is_better": true
|
| 637 |
+
}
|
| 638 |
+
],
|
| 639 |
+
"output_type": "multiple_choice",
|
| 640 |
+
"repeats": 1,
|
| 641 |
+
"should_decontaminate": false,
|
| 642 |
+
"metadata": {
|
| 643 |
+
"version": 0.0
|
| 644 |
+
}
|
| 645 |
+
},
|
| 646 |
+
"mmlu_college_mathematics": {
|
| 647 |
+
"task": "mmlu_college_mathematics",
|
| 648 |
+
"task_alias": "college_mathematics",
|
| 649 |
+
"group": "mmlu_stem",
|
| 650 |
+
"group_alias": "stem",
|
| 651 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 652 |
+
"dataset_name": "college_mathematics",
|
| 653 |
+
"test_split": "test",
|
| 654 |
+
"fewshot_split": "dev",
|
| 655 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 656 |
+
"doc_to_target": "answer",
|
| 657 |
+
"doc_to_choice": [
|
| 658 |
+
"A",
|
| 659 |
+
"B",
|
| 660 |
+
"C",
|
| 661 |
+
"D"
|
| 662 |
+
],
|
| 663 |
+
"description": "The following are multiple choice questions (with answers) about college mathematics.\n\n",
|
| 664 |
+
"target_delimiter": " ",
|
| 665 |
+
"fewshot_delimiter": "\n\n",
|
| 666 |
+
"fewshot_config": {
|
| 667 |
+
"sampler": "first_n"
|
| 668 |
+
},
|
| 669 |
+
"num_fewshot": 5,
|
| 670 |
+
"metric_list": [
|
| 671 |
+
{
|
| 672 |
+
"metric": "acc",
|
| 673 |
+
"aggregation": "mean",
|
| 674 |
+
"higher_is_better": true
|
| 675 |
+
}
|
| 676 |
+
],
|
| 677 |
+
"output_type": "multiple_choice",
|
| 678 |
+
"repeats": 1,
|
| 679 |
+
"should_decontaminate": false,
|
| 680 |
+
"metadata": {
|
| 681 |
+
"version": 0.0
|
| 682 |
+
}
|
| 683 |
+
},
|
| 684 |
+
"mmlu_college_medicine": {
|
| 685 |
+
"task": "mmlu_college_medicine",
|
| 686 |
+
"task_alias": "college_medicine",
|
| 687 |
+
"group": "mmlu_other",
|
| 688 |
+
"group_alias": "other",
|
| 689 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 690 |
+
"dataset_name": "college_medicine",
|
| 691 |
+
"test_split": "test",
|
| 692 |
+
"fewshot_split": "dev",
|
| 693 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 694 |
+
"doc_to_target": "answer",
|
| 695 |
+
"doc_to_choice": [
|
| 696 |
+
"A",
|
| 697 |
+
"B",
|
| 698 |
+
"C",
|
| 699 |
+
"D"
|
| 700 |
+
],
|
| 701 |
+
"description": "The following are multiple choice questions (with answers) about college medicine.\n\n",
|
| 702 |
+
"target_delimiter": " ",
|
| 703 |
+
"fewshot_delimiter": "\n\n",
|
| 704 |
+
"fewshot_config": {
|
| 705 |
+
"sampler": "first_n"
|
| 706 |
+
},
|
| 707 |
+
"num_fewshot": 5,
|
| 708 |
+
"metric_list": [
|
| 709 |
+
{
|
| 710 |
+
"metric": "acc",
|
| 711 |
+
"aggregation": "mean",
|
| 712 |
+
"higher_is_better": true
|
| 713 |
+
}
|
| 714 |
+
],
|
| 715 |
+
"output_type": "multiple_choice",
|
| 716 |
+
"repeats": 1,
|
| 717 |
+
"should_decontaminate": false,
|
| 718 |
+
"metadata": {
|
| 719 |
+
"version": 0.0
|
| 720 |
+
}
|
| 721 |
+
},
|
| 722 |
+
"mmlu_college_physics": {
|
| 723 |
+
"task": "mmlu_college_physics",
|
| 724 |
+
"task_alias": "college_physics",
|
| 725 |
+
"group": "mmlu_stem",
|
| 726 |
+
"group_alias": "stem",
|
| 727 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 728 |
+
"dataset_name": "college_physics",
|
| 729 |
+
"test_split": "test",
|
| 730 |
+
"fewshot_split": "dev",
|
| 731 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 732 |
+
"doc_to_target": "answer",
|
| 733 |
+
"doc_to_choice": [
|
| 734 |
+
"A",
|
| 735 |
+
"B",
|
| 736 |
+
"C",
|
| 737 |
+
"D"
|
| 738 |
+
],
|
| 739 |
+
"description": "The following are multiple choice questions (with answers) about college physics.\n\n",
|
| 740 |
+
"target_delimiter": " ",
|
| 741 |
+
"fewshot_delimiter": "\n\n",
|
| 742 |
+
"fewshot_config": {
|
| 743 |
+
"sampler": "first_n"
|
| 744 |
+
},
|
| 745 |
+
"num_fewshot": 5,
|
| 746 |
+
"metric_list": [
|
| 747 |
+
{
|
| 748 |
+
"metric": "acc",
|
| 749 |
+
"aggregation": "mean",
|
| 750 |
+
"higher_is_better": true
|
| 751 |
+
}
|
| 752 |
+
],
|
| 753 |
+
"output_type": "multiple_choice",
|
| 754 |
+
"repeats": 1,
|
| 755 |
+
"should_decontaminate": false,
|
| 756 |
+
"metadata": {
|
| 757 |
+
"version": 0.0
|
| 758 |
+
}
|
| 759 |
+
},
|
| 760 |
+
"mmlu_computer_security": {
|
| 761 |
+
"task": "mmlu_computer_security",
|
| 762 |
+
"task_alias": "computer_security",
|
| 763 |
+
"group": "mmlu_stem",
|
| 764 |
+
"group_alias": "stem",
|
| 765 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 766 |
+
"dataset_name": "computer_security",
|
| 767 |
+
"test_split": "test",
|
| 768 |
+
"fewshot_split": "dev",
|
| 769 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 770 |
+
"doc_to_target": "answer",
|
| 771 |
+
"doc_to_choice": [
|
| 772 |
+
"A",
|
| 773 |
+
"B",
|
| 774 |
+
"C",
|
| 775 |
+
"D"
|
| 776 |
+
],
|
| 777 |
+
"description": "The following are multiple choice questions (with answers) about computer security.\n\n",
|
| 778 |
+
"target_delimiter": " ",
|
| 779 |
+
"fewshot_delimiter": "\n\n",
|
| 780 |
+
"fewshot_config": {
|
| 781 |
+
"sampler": "first_n"
|
| 782 |
+
},
|
| 783 |
+
"num_fewshot": 5,
|
| 784 |
+
"metric_list": [
|
| 785 |
+
{
|
| 786 |
+
"metric": "acc",
|
| 787 |
+
"aggregation": "mean",
|
| 788 |
+
"higher_is_better": true
|
| 789 |
+
}
|
| 790 |
+
],
|
| 791 |
+
"output_type": "multiple_choice",
|
| 792 |
+
"repeats": 1,
|
| 793 |
+
"should_decontaminate": false,
|
| 794 |
+
"metadata": {
|
| 795 |
+
"version": 0.0
|
| 796 |
+
}
|
| 797 |
+
},
|
| 798 |
+
"mmlu_conceptual_physics": {
|
| 799 |
+
"task": "mmlu_conceptual_physics",
|
| 800 |
+
"task_alias": "conceptual_physics",
|
| 801 |
+
"group": "mmlu_stem",
|
| 802 |
+
"group_alias": "stem",
|
| 803 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 804 |
+
"dataset_name": "conceptual_physics",
|
| 805 |
+
"test_split": "test",
|
| 806 |
+
"fewshot_split": "dev",
|
| 807 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 808 |
+
"doc_to_target": "answer",
|
| 809 |
+
"doc_to_choice": [
|
| 810 |
+
"A",
|
| 811 |
+
"B",
|
| 812 |
+
"C",
|
| 813 |
+
"D"
|
| 814 |
+
],
|
| 815 |
+
"description": "The following are multiple choice questions (with answers) about conceptual physics.\n\n",
|
| 816 |
+
"target_delimiter": " ",
|
| 817 |
+
"fewshot_delimiter": "\n\n",
|
| 818 |
+
"fewshot_config": {
|
| 819 |
+
"sampler": "first_n"
|
| 820 |
+
},
|
| 821 |
+
"num_fewshot": 5,
|
| 822 |
+
"metric_list": [
|
| 823 |
+
{
|
| 824 |
+
"metric": "acc",
|
| 825 |
+
"aggregation": "mean",
|
| 826 |
+
"higher_is_better": true
|
| 827 |
+
}
|
| 828 |
+
],
|
| 829 |
+
"output_type": "multiple_choice",
|
| 830 |
+
"repeats": 1,
|
| 831 |
+
"should_decontaminate": false,
|
| 832 |
+
"metadata": {
|
| 833 |
+
"version": 0.0
|
| 834 |
+
}
|
| 835 |
+
},
|
| 836 |
+
"mmlu_econometrics": {
|
| 837 |
+
"task": "mmlu_econometrics",
|
| 838 |
+
"task_alias": "econometrics",
|
| 839 |
+
"group": "mmlu_social_sciences",
|
| 840 |
+
"group_alias": "social_sciences",
|
| 841 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 842 |
+
"dataset_name": "econometrics",
|
| 843 |
+
"test_split": "test",
|
| 844 |
+
"fewshot_split": "dev",
|
| 845 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 846 |
+
"doc_to_target": "answer",
|
| 847 |
+
"doc_to_choice": [
|
| 848 |
+
"A",
|
| 849 |
+
"B",
|
| 850 |
+
"C",
|
| 851 |
+
"D"
|
| 852 |
+
],
|
| 853 |
+
"description": "The following are multiple choice questions (with answers) about econometrics.\n\n",
|
| 854 |
+
"target_delimiter": " ",
|
| 855 |
+
"fewshot_delimiter": "\n\n",
|
| 856 |
+
"fewshot_config": {
|
| 857 |
+
"sampler": "first_n"
|
| 858 |
+
},
|
| 859 |
+
"num_fewshot": 5,
|
| 860 |
+
"metric_list": [
|
| 861 |
+
{
|
| 862 |
+
"metric": "acc",
|
| 863 |
+
"aggregation": "mean",
|
| 864 |
+
"higher_is_better": true
|
| 865 |
+
}
|
| 866 |
+
],
|
| 867 |
+
"output_type": "multiple_choice",
|
| 868 |
+
"repeats": 1,
|
| 869 |
+
"should_decontaminate": false,
|
| 870 |
+
"metadata": {
|
| 871 |
+
"version": 0.0
|
| 872 |
+
}
|
| 873 |
+
},
|
| 874 |
+
"mmlu_electrical_engineering": {
|
| 875 |
+
"task": "mmlu_electrical_engineering",
|
| 876 |
+
"task_alias": "electrical_engineering",
|
| 877 |
+
"group": "mmlu_stem",
|
| 878 |
+
"group_alias": "stem",
|
| 879 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 880 |
+
"dataset_name": "electrical_engineering",
|
| 881 |
+
"test_split": "test",
|
| 882 |
+
"fewshot_split": "dev",
|
| 883 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 884 |
+
"doc_to_target": "answer",
|
| 885 |
+
"doc_to_choice": [
|
| 886 |
+
"A",
|
| 887 |
+
"B",
|
| 888 |
+
"C",
|
| 889 |
+
"D"
|
| 890 |
+
],
|
| 891 |
+
"description": "The following are multiple choice questions (with answers) about electrical engineering.\n\n",
|
| 892 |
+
"target_delimiter": " ",
|
| 893 |
+
"fewshot_delimiter": "\n\n",
|
| 894 |
+
"fewshot_config": {
|
| 895 |
+
"sampler": "first_n"
|
| 896 |
+
},
|
| 897 |
+
"num_fewshot": 5,
|
| 898 |
+
"metric_list": [
|
| 899 |
+
{
|
| 900 |
+
"metric": "acc",
|
| 901 |
+
"aggregation": "mean",
|
| 902 |
+
"higher_is_better": true
|
| 903 |
+
}
|
| 904 |
+
],
|
| 905 |
+
"output_type": "multiple_choice",
|
| 906 |
+
"repeats": 1,
|
| 907 |
+
"should_decontaminate": false,
|
| 908 |
+
"metadata": {
|
| 909 |
+
"version": 0.0
|
| 910 |
+
}
|
| 911 |
+
},
|
| 912 |
+
"mmlu_elementary_mathematics": {
|
| 913 |
+
"task": "mmlu_elementary_mathematics",
|
| 914 |
+
"task_alias": "elementary_mathematics",
|
| 915 |
+
"group": "mmlu_stem",
|
| 916 |
+
"group_alias": "stem",
|
| 917 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 918 |
+
"dataset_name": "elementary_mathematics",
|
| 919 |
+
"test_split": "test",
|
| 920 |
+
"fewshot_split": "dev",
|
| 921 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 922 |
+
"doc_to_target": "answer",
|
| 923 |
+
"doc_to_choice": [
|
| 924 |
+
"A",
|
| 925 |
+
"B",
|
| 926 |
+
"C",
|
| 927 |
+
"D"
|
| 928 |
+
],
|
| 929 |
+
"description": "The following are multiple choice questions (with answers) about elementary mathematics.\n\n",
|
| 930 |
+
"target_delimiter": " ",
|
| 931 |
+
"fewshot_delimiter": "\n\n",
|
| 932 |
+
"fewshot_config": {
|
| 933 |
+
"sampler": "first_n"
|
| 934 |
+
},
|
| 935 |
+
"num_fewshot": 5,
|
| 936 |
+
"metric_list": [
|
| 937 |
+
{
|
| 938 |
+
"metric": "acc",
|
| 939 |
+
"aggregation": "mean",
|
| 940 |
+
"higher_is_better": true
|
| 941 |
+
}
|
| 942 |
+
],
|
| 943 |
+
"output_type": "multiple_choice",
|
| 944 |
+
"repeats": 1,
|
| 945 |
+
"should_decontaminate": false,
|
| 946 |
+
"metadata": {
|
| 947 |
+
"version": 0.0
|
| 948 |
+
}
|
| 949 |
+
},
|
| 950 |
+
"mmlu_formal_logic": {
|
| 951 |
+
"task": "mmlu_formal_logic",
|
| 952 |
+
"task_alias": "formal_logic",
|
| 953 |
+
"group": "mmlu_humanities",
|
| 954 |
+
"group_alias": "humanities",
|
| 955 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 956 |
+
"dataset_name": "formal_logic",
|
| 957 |
+
"test_split": "test",
|
| 958 |
+
"fewshot_split": "dev",
|
| 959 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 960 |
+
"doc_to_target": "answer",
|
| 961 |
+
"doc_to_choice": [
|
| 962 |
+
"A",
|
| 963 |
+
"B",
|
| 964 |
+
"C",
|
| 965 |
+
"D"
|
| 966 |
+
],
|
| 967 |
+
"description": "The following are multiple choice questions (with answers) about formal logic.\n\n",
|
| 968 |
+
"target_delimiter": " ",
|
| 969 |
+
"fewshot_delimiter": "\n\n",
|
| 970 |
+
"fewshot_config": {
|
| 971 |
+
"sampler": "first_n"
|
| 972 |
+
},
|
| 973 |
+
"num_fewshot": 5,
|
| 974 |
+
"metric_list": [
|
| 975 |
+
{
|
| 976 |
+
"metric": "acc",
|
| 977 |
+
"aggregation": "mean",
|
| 978 |
+
"higher_is_better": true
|
| 979 |
+
}
|
| 980 |
+
],
|
| 981 |
+
"output_type": "multiple_choice",
|
| 982 |
+
"repeats": 1,
|
| 983 |
+
"should_decontaminate": false,
|
| 984 |
+
"metadata": {
|
| 985 |
+
"version": 0.0
|
| 986 |
+
}
|
| 987 |
+
},
|
| 988 |
+
"mmlu_global_facts": {
|
| 989 |
+
"task": "mmlu_global_facts",
|
| 990 |
+
"task_alias": "global_facts",
|
| 991 |
+
"group": "mmlu_other",
|
| 992 |
+
"group_alias": "other",
|
| 993 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 994 |
+
"dataset_name": "global_facts",
|
| 995 |
+
"test_split": "test",
|
| 996 |
+
"fewshot_split": "dev",
|
| 997 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 998 |
+
"doc_to_target": "answer",
|
| 999 |
+
"doc_to_choice": [
|
| 1000 |
+
"A",
|
| 1001 |
+
"B",
|
| 1002 |
+
"C",
|
| 1003 |
+
"D"
|
| 1004 |
+
],
|
| 1005 |
+
"description": "The following are multiple choice questions (with answers) about global facts.\n\n",
|
| 1006 |
+
"target_delimiter": " ",
|
| 1007 |
+
"fewshot_delimiter": "\n\n",
|
| 1008 |
+
"fewshot_config": {
|
| 1009 |
+
"sampler": "first_n"
|
| 1010 |
+
},
|
| 1011 |
+
"num_fewshot": 5,
|
| 1012 |
+
"metric_list": [
|
| 1013 |
+
{
|
| 1014 |
+
"metric": "acc",
|
| 1015 |
+
"aggregation": "mean",
|
| 1016 |
+
"higher_is_better": true
|
| 1017 |
+
}
|
| 1018 |
+
],
|
| 1019 |
+
"output_type": "multiple_choice",
|
| 1020 |
+
"repeats": 1,
|
| 1021 |
+
"should_decontaminate": false,
|
| 1022 |
+
"metadata": {
|
| 1023 |
+
"version": 0.0
|
| 1024 |
+
}
|
| 1025 |
+
},
|
| 1026 |
+
"mmlu_high_school_biology": {
|
| 1027 |
+
"task": "mmlu_high_school_biology",
|
| 1028 |
+
"task_alias": "high_school_biology",
|
| 1029 |
+
"group": "mmlu_stem",
|
| 1030 |
+
"group_alias": "stem",
|
| 1031 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1032 |
+
"dataset_name": "high_school_biology",
|
| 1033 |
+
"test_split": "test",
|
| 1034 |
+
"fewshot_split": "dev",
|
| 1035 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1036 |
+
"doc_to_target": "answer",
|
| 1037 |
+
"doc_to_choice": [
|
| 1038 |
+
"A",
|
| 1039 |
+
"B",
|
| 1040 |
+
"C",
|
| 1041 |
+
"D"
|
| 1042 |
+
],
|
| 1043 |
+
"description": "The following are multiple choice questions (with answers) about high school biology.\n\n",
|
| 1044 |
+
"target_delimiter": " ",
|
| 1045 |
+
"fewshot_delimiter": "\n\n",
|
| 1046 |
+
"fewshot_config": {
|
| 1047 |
+
"sampler": "first_n"
|
| 1048 |
+
},
|
| 1049 |
+
"num_fewshot": 5,
|
| 1050 |
+
"metric_list": [
|
| 1051 |
+
{
|
| 1052 |
+
"metric": "acc",
|
| 1053 |
+
"aggregation": "mean",
|
| 1054 |
+
"higher_is_better": true
|
| 1055 |
+
}
|
| 1056 |
+
],
|
| 1057 |
+
"output_type": "multiple_choice",
|
| 1058 |
+
"repeats": 1,
|
| 1059 |
+
"should_decontaminate": false,
|
| 1060 |
+
"metadata": {
|
| 1061 |
+
"version": 0.0
|
| 1062 |
+
}
|
| 1063 |
+
},
|
| 1064 |
+
"mmlu_high_school_chemistry": {
|
| 1065 |
+
"task": "mmlu_high_school_chemistry",
|
| 1066 |
+
"task_alias": "high_school_chemistry",
|
| 1067 |
+
"group": "mmlu_stem",
|
| 1068 |
+
"group_alias": "stem",
|
| 1069 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1070 |
+
"dataset_name": "high_school_chemistry",
|
| 1071 |
+
"test_split": "test",
|
| 1072 |
+
"fewshot_split": "dev",
|
| 1073 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1074 |
+
"doc_to_target": "answer",
|
| 1075 |
+
"doc_to_choice": [
|
| 1076 |
+
"A",
|
| 1077 |
+
"B",
|
| 1078 |
+
"C",
|
| 1079 |
+
"D"
|
| 1080 |
+
],
|
| 1081 |
+
"description": "The following are multiple choice questions (with answers) about high school chemistry.\n\n",
|
| 1082 |
+
"target_delimiter": " ",
|
| 1083 |
+
"fewshot_delimiter": "\n\n",
|
| 1084 |
+
"fewshot_config": {
|
| 1085 |
+
"sampler": "first_n"
|
| 1086 |
+
},
|
| 1087 |
+
"num_fewshot": 5,
|
| 1088 |
+
"metric_list": [
|
| 1089 |
+
{
|
| 1090 |
+
"metric": "acc",
|
| 1091 |
+
"aggregation": "mean",
|
| 1092 |
+
"higher_is_better": true
|
| 1093 |
+
}
|
| 1094 |
+
],
|
| 1095 |
+
"output_type": "multiple_choice",
|
| 1096 |
+
"repeats": 1,
|
| 1097 |
+
"should_decontaminate": false,
|
| 1098 |
+
"metadata": {
|
| 1099 |
+
"version": 0.0
|
| 1100 |
+
}
|
| 1101 |
+
},
|
| 1102 |
+
"mmlu_high_school_computer_science": {
|
| 1103 |
+
"task": "mmlu_high_school_computer_science",
|
| 1104 |
+
"task_alias": "high_school_computer_science",
|
| 1105 |
+
"group": "mmlu_stem",
|
| 1106 |
+
"group_alias": "stem",
|
| 1107 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1108 |
+
"dataset_name": "high_school_computer_science",
|
| 1109 |
+
"test_split": "test",
|
| 1110 |
+
"fewshot_split": "dev",
|
| 1111 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1112 |
+
"doc_to_target": "answer",
|
| 1113 |
+
"doc_to_choice": [
|
| 1114 |
+
"A",
|
| 1115 |
+
"B",
|
| 1116 |
+
"C",
|
| 1117 |
+
"D"
|
| 1118 |
+
],
|
| 1119 |
+
"description": "The following are multiple choice questions (with answers) about high school computer science.\n\n",
|
| 1120 |
+
"target_delimiter": " ",
|
| 1121 |
+
"fewshot_delimiter": "\n\n",
|
| 1122 |
+
"fewshot_config": {
|
| 1123 |
+
"sampler": "first_n"
|
| 1124 |
+
},
|
| 1125 |
+
"num_fewshot": 5,
|
| 1126 |
+
"metric_list": [
|
| 1127 |
+
{
|
| 1128 |
+
"metric": "acc",
|
| 1129 |
+
"aggregation": "mean",
|
| 1130 |
+
"higher_is_better": true
|
| 1131 |
+
}
|
| 1132 |
+
],
|
| 1133 |
+
"output_type": "multiple_choice",
|
| 1134 |
+
"repeats": 1,
|
| 1135 |
+
"should_decontaminate": false,
|
| 1136 |
+
"metadata": {
|
| 1137 |
+
"version": 0.0
|
| 1138 |
+
}
|
| 1139 |
+
},
|
| 1140 |
+
"mmlu_high_school_european_history": {
|
| 1141 |
+
"task": "mmlu_high_school_european_history",
|
| 1142 |
+
"task_alias": "high_school_european_history",
|
| 1143 |
+
"group": "mmlu_humanities",
|
| 1144 |
+
"group_alias": "humanities",
|
| 1145 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1146 |
+
"dataset_name": "high_school_european_history",
|
| 1147 |
+
"test_split": "test",
|
| 1148 |
+
"fewshot_split": "dev",
|
| 1149 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1150 |
+
"doc_to_target": "answer",
|
| 1151 |
+
"doc_to_choice": [
|
| 1152 |
+
"A",
|
| 1153 |
+
"B",
|
| 1154 |
+
"C",
|
| 1155 |
+
"D"
|
| 1156 |
+
],
|
| 1157 |
+
"description": "The following are multiple choice questions (with answers) about high school european history.\n\n",
|
| 1158 |
+
"target_delimiter": " ",
|
| 1159 |
+
"fewshot_delimiter": "\n\n",
|
| 1160 |
+
"fewshot_config": {
|
| 1161 |
+
"sampler": "first_n"
|
| 1162 |
+
},
|
| 1163 |
+
"num_fewshot": 5,
|
| 1164 |
+
"metric_list": [
|
| 1165 |
+
{
|
| 1166 |
+
"metric": "acc",
|
| 1167 |
+
"aggregation": "mean",
|
| 1168 |
+
"higher_is_better": true
|
| 1169 |
+
}
|
| 1170 |
+
],
|
| 1171 |
+
"output_type": "multiple_choice",
|
| 1172 |
+
"repeats": 1,
|
| 1173 |
+
"should_decontaminate": false,
|
| 1174 |
+
"metadata": {
|
| 1175 |
+
"version": 0.0
|
| 1176 |
+
}
|
| 1177 |
+
},
|
| 1178 |
+
"mmlu_high_school_geography": {
|
| 1179 |
+
"task": "mmlu_high_school_geography",
|
| 1180 |
+
"task_alias": "high_school_geography",
|
| 1181 |
+
"group": "mmlu_social_sciences",
|
| 1182 |
+
"group_alias": "social_sciences",
|
| 1183 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1184 |
+
"dataset_name": "high_school_geography",
|
| 1185 |
+
"test_split": "test",
|
| 1186 |
+
"fewshot_split": "dev",
|
| 1187 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1188 |
+
"doc_to_target": "answer",
|
| 1189 |
+
"doc_to_choice": [
|
| 1190 |
+
"A",
|
| 1191 |
+
"B",
|
| 1192 |
+
"C",
|
| 1193 |
+
"D"
|
| 1194 |
+
],
|
| 1195 |
+
"description": "The following are multiple choice questions (with answers) about high school geography.\n\n",
|
| 1196 |
+
"target_delimiter": " ",
|
| 1197 |
+
"fewshot_delimiter": "\n\n",
|
| 1198 |
+
"fewshot_config": {
|
| 1199 |
+
"sampler": "first_n"
|
| 1200 |
+
},
|
| 1201 |
+
"num_fewshot": 5,
|
| 1202 |
+
"metric_list": [
|
| 1203 |
+
{
|
| 1204 |
+
"metric": "acc",
|
| 1205 |
+
"aggregation": "mean",
|
| 1206 |
+
"higher_is_better": true
|
| 1207 |
+
}
|
| 1208 |
+
],
|
| 1209 |
+
"output_type": "multiple_choice",
|
| 1210 |
+
"repeats": 1,
|
| 1211 |
+
"should_decontaminate": false,
|
| 1212 |
+
"metadata": {
|
| 1213 |
+
"version": 0.0
|
| 1214 |
+
}
|
| 1215 |
+
},
|
| 1216 |
+
"mmlu_high_school_government_and_politics": {
|
| 1217 |
+
"task": "mmlu_high_school_government_and_politics",
|
| 1218 |
+
"task_alias": "high_school_government_and_politics",
|
| 1219 |
+
"group": "mmlu_social_sciences",
|
| 1220 |
+
"group_alias": "social_sciences",
|
| 1221 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1222 |
+
"dataset_name": "high_school_government_and_politics",
|
| 1223 |
+
"test_split": "test",
|
| 1224 |
+
"fewshot_split": "dev",
|
| 1225 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1226 |
+
"doc_to_target": "answer",
|
| 1227 |
+
"doc_to_choice": [
|
| 1228 |
+
"A",
|
| 1229 |
+
"B",
|
| 1230 |
+
"C",
|
| 1231 |
+
"D"
|
| 1232 |
+
],
|
| 1233 |
+
"description": "The following are multiple choice questions (with answers) about high school government and politics.\n\n",
|
| 1234 |
+
"target_delimiter": " ",
|
| 1235 |
+
"fewshot_delimiter": "\n\n",
|
| 1236 |
+
"fewshot_config": {
|
| 1237 |
+
"sampler": "first_n"
|
| 1238 |
+
},
|
| 1239 |
+
"num_fewshot": 5,
|
| 1240 |
+
"metric_list": [
|
| 1241 |
+
{
|
| 1242 |
+
"metric": "acc",
|
| 1243 |
+
"aggregation": "mean",
|
| 1244 |
+
"higher_is_better": true
|
| 1245 |
+
}
|
| 1246 |
+
],
|
| 1247 |
+
"output_type": "multiple_choice",
|
| 1248 |
+
"repeats": 1,
|
| 1249 |
+
"should_decontaminate": false,
|
| 1250 |
+
"metadata": {
|
| 1251 |
+
"version": 0.0
|
| 1252 |
+
}
|
| 1253 |
+
},
|
| 1254 |
+
"mmlu_high_school_macroeconomics": {
|
| 1255 |
+
"task": "mmlu_high_school_macroeconomics",
|
| 1256 |
+
"task_alias": "high_school_macroeconomics",
|
| 1257 |
+
"group": "mmlu_social_sciences",
|
| 1258 |
+
"group_alias": "social_sciences",
|
| 1259 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1260 |
+
"dataset_name": "high_school_macroeconomics",
|
| 1261 |
+
"test_split": "test",
|
| 1262 |
+
"fewshot_split": "dev",
|
| 1263 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1264 |
+
"doc_to_target": "answer",
|
| 1265 |
+
"doc_to_choice": [
|
| 1266 |
+
"A",
|
| 1267 |
+
"B",
|
| 1268 |
+
"C",
|
| 1269 |
+
"D"
|
| 1270 |
+
],
|
| 1271 |
+
"description": "The following are multiple choice questions (with answers) about high school macroeconomics.\n\n",
|
| 1272 |
+
"target_delimiter": " ",
|
| 1273 |
+
"fewshot_delimiter": "\n\n",
|
| 1274 |
+
"fewshot_config": {
|
| 1275 |
+
"sampler": "first_n"
|
| 1276 |
+
},
|
| 1277 |
+
"num_fewshot": 5,
|
| 1278 |
+
"metric_list": [
|
| 1279 |
+
{
|
| 1280 |
+
"metric": "acc",
|
| 1281 |
+
"aggregation": "mean",
|
| 1282 |
+
"higher_is_better": true
|
| 1283 |
+
}
|
| 1284 |
+
],
|
| 1285 |
+
"output_type": "multiple_choice",
|
| 1286 |
+
"repeats": 1,
|
| 1287 |
+
"should_decontaminate": false,
|
| 1288 |
+
"metadata": {
|
| 1289 |
+
"version": 0.0
|
| 1290 |
+
}
|
| 1291 |
+
},
|
| 1292 |
+
"mmlu_high_school_mathematics": {
|
| 1293 |
+
"task": "mmlu_high_school_mathematics",
|
| 1294 |
+
"task_alias": "high_school_mathematics",
|
| 1295 |
+
"group": "mmlu_stem",
|
| 1296 |
+
"group_alias": "stem",
|
| 1297 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1298 |
+
"dataset_name": "high_school_mathematics",
|
| 1299 |
+
"test_split": "test",
|
| 1300 |
+
"fewshot_split": "dev",
|
| 1301 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1302 |
+
"doc_to_target": "answer",
|
| 1303 |
+
"doc_to_choice": [
|
| 1304 |
+
"A",
|
| 1305 |
+
"B",
|
| 1306 |
+
"C",
|
| 1307 |
+
"D"
|
| 1308 |
+
],
|
| 1309 |
+
"description": "The following are multiple choice questions (with answers) about high school mathematics.\n\n",
|
| 1310 |
+
"target_delimiter": " ",
|
| 1311 |
+
"fewshot_delimiter": "\n\n",
|
| 1312 |
+
"fewshot_config": {
|
| 1313 |
+
"sampler": "first_n"
|
| 1314 |
+
},
|
| 1315 |
+
"num_fewshot": 5,
|
| 1316 |
+
"metric_list": [
|
| 1317 |
+
{
|
| 1318 |
+
"metric": "acc",
|
| 1319 |
+
"aggregation": "mean",
|
| 1320 |
+
"higher_is_better": true
|
| 1321 |
+
}
|
| 1322 |
+
],
|
| 1323 |
+
"output_type": "multiple_choice",
|
| 1324 |
+
"repeats": 1,
|
| 1325 |
+
"should_decontaminate": false,
|
| 1326 |
+
"metadata": {
|
| 1327 |
+
"version": 0.0
|
| 1328 |
+
}
|
| 1329 |
+
},
|
| 1330 |
+
"mmlu_high_school_microeconomics": {
|
| 1331 |
+
"task": "mmlu_high_school_microeconomics",
|
| 1332 |
+
"task_alias": "high_school_microeconomics",
|
| 1333 |
+
"group": "mmlu_social_sciences",
|
| 1334 |
+
"group_alias": "social_sciences",
|
| 1335 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1336 |
+
"dataset_name": "high_school_microeconomics",
|
| 1337 |
+
"test_split": "test",
|
| 1338 |
+
"fewshot_split": "dev",
|
| 1339 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1340 |
+
"doc_to_target": "answer",
|
| 1341 |
+
"doc_to_choice": [
|
| 1342 |
+
"A",
|
| 1343 |
+
"B",
|
| 1344 |
+
"C",
|
| 1345 |
+
"D"
|
| 1346 |
+
],
|
| 1347 |
+
"description": "The following are multiple choice questions (with answers) about high school microeconomics.\n\n",
|
| 1348 |
+
"target_delimiter": " ",
|
| 1349 |
+
"fewshot_delimiter": "\n\n",
|
| 1350 |
+
"fewshot_config": {
|
| 1351 |
+
"sampler": "first_n"
|
| 1352 |
+
},
|
| 1353 |
+
"num_fewshot": 5,
|
| 1354 |
+
"metric_list": [
|
| 1355 |
+
{
|
| 1356 |
+
"metric": "acc",
|
| 1357 |
+
"aggregation": "mean",
|
| 1358 |
+
"higher_is_better": true
|
| 1359 |
+
}
|
| 1360 |
+
],
|
| 1361 |
+
"output_type": "multiple_choice",
|
| 1362 |
+
"repeats": 1,
|
| 1363 |
+
"should_decontaminate": false,
|
| 1364 |
+
"metadata": {
|
| 1365 |
+
"version": 0.0
|
| 1366 |
+
}
|
| 1367 |
+
},
|
| 1368 |
+
"mmlu_high_school_physics": {
|
| 1369 |
+
"task": "mmlu_high_school_physics",
|
| 1370 |
+
"task_alias": "high_school_physics",
|
| 1371 |
+
"group": "mmlu_stem",
|
| 1372 |
+
"group_alias": "stem",
|
| 1373 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1374 |
+
"dataset_name": "high_school_physics",
|
| 1375 |
+
"test_split": "test",
|
| 1376 |
+
"fewshot_split": "dev",
|
| 1377 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1378 |
+
"doc_to_target": "answer",
|
| 1379 |
+
"doc_to_choice": [
|
| 1380 |
+
"A",
|
| 1381 |
+
"B",
|
| 1382 |
+
"C",
|
| 1383 |
+
"D"
|
| 1384 |
+
],
|
| 1385 |
+
"description": "The following are multiple choice questions (with answers) about high school physics.\n\n",
|
| 1386 |
+
"target_delimiter": " ",
|
| 1387 |
+
"fewshot_delimiter": "\n\n",
|
| 1388 |
+
"fewshot_config": {
|
| 1389 |
+
"sampler": "first_n"
|
| 1390 |
+
},
|
| 1391 |
+
"num_fewshot": 5,
|
| 1392 |
+
"metric_list": [
|
| 1393 |
+
{
|
| 1394 |
+
"metric": "acc",
|
| 1395 |
+
"aggregation": "mean",
|
| 1396 |
+
"higher_is_better": true
|
| 1397 |
+
}
|
| 1398 |
+
],
|
| 1399 |
+
"output_type": "multiple_choice",
|
| 1400 |
+
"repeats": 1,
|
| 1401 |
+
"should_decontaminate": false,
|
| 1402 |
+
"metadata": {
|
| 1403 |
+
"version": 0.0
|
| 1404 |
+
}
|
| 1405 |
+
},
|
| 1406 |
+
"mmlu_high_school_psychology": {
|
| 1407 |
+
"task": "mmlu_high_school_psychology",
|
| 1408 |
+
"task_alias": "high_school_psychology",
|
| 1409 |
+
"group": "mmlu_social_sciences",
|
| 1410 |
+
"group_alias": "social_sciences",
|
| 1411 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1412 |
+
"dataset_name": "high_school_psychology",
|
| 1413 |
+
"test_split": "test",
|
| 1414 |
+
"fewshot_split": "dev",
|
| 1415 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1416 |
+
"doc_to_target": "answer",
|
| 1417 |
+
"doc_to_choice": [
|
| 1418 |
+
"A",
|
| 1419 |
+
"B",
|
| 1420 |
+
"C",
|
| 1421 |
+
"D"
|
| 1422 |
+
],
|
| 1423 |
+
"description": "The following are multiple choice questions (with answers) about high school psychology.\n\n",
|
| 1424 |
+
"target_delimiter": " ",
|
| 1425 |
+
"fewshot_delimiter": "\n\n",
|
| 1426 |
+
"fewshot_config": {
|
| 1427 |
+
"sampler": "first_n"
|
| 1428 |
+
},
|
| 1429 |
+
"num_fewshot": 5,
|
| 1430 |
+
"metric_list": [
|
| 1431 |
+
{
|
| 1432 |
+
"metric": "acc",
|
| 1433 |
+
"aggregation": "mean",
|
| 1434 |
+
"higher_is_better": true
|
| 1435 |
+
}
|
| 1436 |
+
],
|
| 1437 |
+
"output_type": "multiple_choice",
|
| 1438 |
+
"repeats": 1,
|
| 1439 |
+
"should_decontaminate": false,
|
| 1440 |
+
"metadata": {
|
| 1441 |
+
"version": 0.0
|
| 1442 |
+
}
|
| 1443 |
+
},
|
| 1444 |
+
"mmlu_high_school_statistics": {
|
| 1445 |
+
"task": "mmlu_high_school_statistics",
|
| 1446 |
+
"task_alias": "high_school_statistics",
|
| 1447 |
+
"group": "mmlu_stem",
|
| 1448 |
+
"group_alias": "stem",
|
| 1449 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1450 |
+
"dataset_name": "high_school_statistics",
|
| 1451 |
+
"test_split": "test",
|
| 1452 |
+
"fewshot_split": "dev",
|
| 1453 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1454 |
+
"doc_to_target": "answer",
|
| 1455 |
+
"doc_to_choice": [
|
| 1456 |
+
"A",
|
| 1457 |
+
"B",
|
| 1458 |
+
"C",
|
| 1459 |
+
"D"
|
| 1460 |
+
],
|
| 1461 |
+
"description": "The following are multiple choice questions (with answers) about high school statistics.\n\n",
|
| 1462 |
+
"target_delimiter": " ",
|
| 1463 |
+
"fewshot_delimiter": "\n\n",
|
| 1464 |
+
"fewshot_config": {
|
| 1465 |
+
"sampler": "first_n"
|
| 1466 |
+
},
|
| 1467 |
+
"num_fewshot": 5,
|
| 1468 |
+
"metric_list": [
|
| 1469 |
+
{
|
| 1470 |
+
"metric": "acc",
|
| 1471 |
+
"aggregation": "mean",
|
| 1472 |
+
"higher_is_better": true
|
| 1473 |
+
}
|
| 1474 |
+
],
|
| 1475 |
+
"output_type": "multiple_choice",
|
| 1476 |
+
"repeats": 1,
|
| 1477 |
+
"should_decontaminate": false,
|
| 1478 |
+
"metadata": {
|
| 1479 |
+
"version": 0.0
|
| 1480 |
+
}
|
| 1481 |
+
},
|
| 1482 |
+
"mmlu_high_school_us_history": {
|
| 1483 |
+
"task": "mmlu_high_school_us_history",
|
| 1484 |
+
"task_alias": "high_school_us_history",
|
| 1485 |
+
"group": "mmlu_humanities",
|
| 1486 |
+
"group_alias": "humanities",
|
| 1487 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1488 |
+
"dataset_name": "high_school_us_history",
|
| 1489 |
+
"test_split": "test",
|
| 1490 |
+
"fewshot_split": "dev",
|
| 1491 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1492 |
+
"doc_to_target": "answer",
|
| 1493 |
+
"doc_to_choice": [
|
| 1494 |
+
"A",
|
| 1495 |
+
"B",
|
| 1496 |
+
"C",
|
| 1497 |
+
"D"
|
| 1498 |
+
],
|
| 1499 |
+
"description": "The following are multiple choice questions (with answers) about high school us history.\n\n",
|
| 1500 |
+
"target_delimiter": " ",
|
| 1501 |
+
"fewshot_delimiter": "\n\n",
|
| 1502 |
+
"fewshot_config": {
|
| 1503 |
+
"sampler": "first_n"
|
| 1504 |
+
},
|
| 1505 |
+
"num_fewshot": 5,
|
| 1506 |
+
"metric_list": [
|
| 1507 |
+
{
|
| 1508 |
+
"metric": "acc",
|
| 1509 |
+
"aggregation": "mean",
|
| 1510 |
+
"higher_is_better": true
|
| 1511 |
+
}
|
| 1512 |
+
],
|
| 1513 |
+
"output_type": "multiple_choice",
|
| 1514 |
+
"repeats": 1,
|
| 1515 |
+
"should_decontaminate": false,
|
| 1516 |
+
"metadata": {
|
| 1517 |
+
"version": 0.0
|
| 1518 |
+
}
|
| 1519 |
+
},
|
| 1520 |
+
"mmlu_high_school_world_history": {
|
| 1521 |
+
"task": "mmlu_high_school_world_history",
|
| 1522 |
+
"task_alias": "high_school_world_history",
|
| 1523 |
+
"group": "mmlu_humanities",
|
| 1524 |
+
"group_alias": "humanities",
|
| 1525 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1526 |
+
"dataset_name": "high_school_world_history",
|
| 1527 |
+
"test_split": "test",
|
| 1528 |
+
"fewshot_split": "dev",
|
| 1529 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1530 |
+
"doc_to_target": "answer",
|
| 1531 |
+
"doc_to_choice": [
|
| 1532 |
+
"A",
|
| 1533 |
+
"B",
|
| 1534 |
+
"C",
|
| 1535 |
+
"D"
|
| 1536 |
+
],
|
| 1537 |
+
"description": "The following are multiple choice questions (with answers) about high school world history.\n\n",
|
| 1538 |
+
"target_delimiter": " ",
|
| 1539 |
+
"fewshot_delimiter": "\n\n",
|
| 1540 |
+
"fewshot_config": {
|
| 1541 |
+
"sampler": "first_n"
|
| 1542 |
+
},
|
| 1543 |
+
"num_fewshot": 5,
|
| 1544 |
+
"metric_list": [
|
| 1545 |
+
{
|
| 1546 |
+
"metric": "acc",
|
| 1547 |
+
"aggregation": "mean",
|
| 1548 |
+
"higher_is_better": true
|
| 1549 |
+
}
|
| 1550 |
+
],
|
| 1551 |
+
"output_type": "multiple_choice",
|
| 1552 |
+
"repeats": 1,
|
| 1553 |
+
"should_decontaminate": false,
|
| 1554 |
+
"metadata": {
|
| 1555 |
+
"version": 0.0
|
| 1556 |
+
}
|
| 1557 |
+
},
|
| 1558 |
+
"mmlu_human_aging": {
|
| 1559 |
+
"task": "mmlu_human_aging",
|
| 1560 |
+
"task_alias": "human_aging",
|
| 1561 |
+
"group": "mmlu_other",
|
| 1562 |
+
"group_alias": "other",
|
| 1563 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1564 |
+
"dataset_name": "human_aging",
|
| 1565 |
+
"test_split": "test",
|
| 1566 |
+
"fewshot_split": "dev",
|
| 1567 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1568 |
+
"doc_to_target": "answer",
|
| 1569 |
+
"doc_to_choice": [
|
| 1570 |
+
"A",
|
| 1571 |
+
"B",
|
| 1572 |
+
"C",
|
| 1573 |
+
"D"
|
| 1574 |
+
],
|
| 1575 |
+
"description": "The following are multiple choice questions (with answers) about human aging.\n\n",
|
| 1576 |
+
"target_delimiter": " ",
|
| 1577 |
+
"fewshot_delimiter": "\n\n",
|
| 1578 |
+
"fewshot_config": {
|
| 1579 |
+
"sampler": "first_n"
|
| 1580 |
+
},
|
| 1581 |
+
"num_fewshot": 5,
|
| 1582 |
+
"metric_list": [
|
| 1583 |
+
{
|
| 1584 |
+
"metric": "acc",
|
| 1585 |
+
"aggregation": "mean",
|
| 1586 |
+
"higher_is_better": true
|
| 1587 |
+
}
|
| 1588 |
+
],
|
| 1589 |
+
"output_type": "multiple_choice",
|
| 1590 |
+
"repeats": 1,
|
| 1591 |
+
"should_decontaminate": false,
|
| 1592 |
+
"metadata": {
|
| 1593 |
+
"version": 0.0
|
| 1594 |
+
}
|
| 1595 |
+
},
|
| 1596 |
+
"mmlu_human_sexuality": {
|
| 1597 |
+
"task": "mmlu_human_sexuality",
|
| 1598 |
+
"task_alias": "human_sexuality",
|
| 1599 |
+
"group": "mmlu_social_sciences",
|
| 1600 |
+
"group_alias": "social_sciences",
|
| 1601 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1602 |
+
"dataset_name": "human_sexuality",
|
| 1603 |
+
"test_split": "test",
|
| 1604 |
+
"fewshot_split": "dev",
|
| 1605 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1606 |
+
"doc_to_target": "answer",
|
| 1607 |
+
"doc_to_choice": [
|
| 1608 |
+
"A",
|
| 1609 |
+
"B",
|
| 1610 |
+
"C",
|
| 1611 |
+
"D"
|
| 1612 |
+
],
|
| 1613 |
+
"description": "The following are multiple choice questions (with answers) about human sexuality.\n\n",
|
| 1614 |
+
"target_delimiter": " ",
|
| 1615 |
+
"fewshot_delimiter": "\n\n",
|
| 1616 |
+
"fewshot_config": {
|
| 1617 |
+
"sampler": "first_n"
|
| 1618 |
+
},
|
| 1619 |
+
"num_fewshot": 5,
|
| 1620 |
+
"metric_list": [
|
| 1621 |
+
{
|
| 1622 |
+
"metric": "acc",
|
| 1623 |
+
"aggregation": "mean",
|
| 1624 |
+
"higher_is_better": true
|
| 1625 |
+
}
|
| 1626 |
+
],
|
| 1627 |
+
"output_type": "multiple_choice",
|
| 1628 |
+
"repeats": 1,
|
| 1629 |
+
"should_decontaminate": false,
|
| 1630 |
+
"metadata": {
|
| 1631 |
+
"version": 0.0
|
| 1632 |
+
}
|
| 1633 |
+
},
|
| 1634 |
+
"mmlu_international_law": {
|
| 1635 |
+
"task": "mmlu_international_law",
|
| 1636 |
+
"task_alias": "international_law",
|
| 1637 |
+
"group": "mmlu_humanities",
|
| 1638 |
+
"group_alias": "humanities",
|
| 1639 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1640 |
+
"dataset_name": "international_law",
|
| 1641 |
+
"test_split": "test",
|
| 1642 |
+
"fewshot_split": "dev",
|
| 1643 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1644 |
+
"doc_to_target": "answer",
|
| 1645 |
+
"doc_to_choice": [
|
| 1646 |
+
"A",
|
| 1647 |
+
"B",
|
| 1648 |
+
"C",
|
| 1649 |
+
"D"
|
| 1650 |
+
],
|
| 1651 |
+
"description": "The following are multiple choice questions (with answers) about international law.\n\n",
|
| 1652 |
+
"target_delimiter": " ",
|
| 1653 |
+
"fewshot_delimiter": "\n\n",
|
| 1654 |
+
"fewshot_config": {
|
| 1655 |
+
"sampler": "first_n"
|
| 1656 |
+
},
|
| 1657 |
+
"num_fewshot": 5,
|
| 1658 |
+
"metric_list": [
|
| 1659 |
+
{
|
| 1660 |
+
"metric": "acc",
|
| 1661 |
+
"aggregation": "mean",
|
| 1662 |
+
"higher_is_better": true
|
| 1663 |
+
}
|
| 1664 |
+
],
|
| 1665 |
+
"output_type": "multiple_choice",
|
| 1666 |
+
"repeats": 1,
|
| 1667 |
+
"should_decontaminate": false,
|
| 1668 |
+
"metadata": {
|
| 1669 |
+
"version": 0.0
|
| 1670 |
+
}
|
| 1671 |
+
},
|
| 1672 |
+
"mmlu_jurisprudence": {
|
| 1673 |
+
"task": "mmlu_jurisprudence",
|
| 1674 |
+
"task_alias": "jurisprudence",
|
| 1675 |
+
"group": "mmlu_humanities",
|
| 1676 |
+
"group_alias": "humanities",
|
| 1677 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1678 |
+
"dataset_name": "jurisprudence",
|
| 1679 |
+
"test_split": "test",
|
| 1680 |
+
"fewshot_split": "dev",
|
| 1681 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1682 |
+
"doc_to_target": "answer",
|
| 1683 |
+
"doc_to_choice": [
|
| 1684 |
+
"A",
|
| 1685 |
+
"B",
|
| 1686 |
+
"C",
|
| 1687 |
+
"D"
|
| 1688 |
+
],
|
| 1689 |
+
"description": "The following are multiple choice questions (with answers) about jurisprudence.\n\n",
|
| 1690 |
+
"target_delimiter": " ",
|
| 1691 |
+
"fewshot_delimiter": "\n\n",
|
| 1692 |
+
"fewshot_config": {
|
| 1693 |
+
"sampler": "first_n"
|
| 1694 |
+
},
|
| 1695 |
+
"num_fewshot": 5,
|
| 1696 |
+
"metric_list": [
|
| 1697 |
+
{
|
| 1698 |
+
"metric": "acc",
|
| 1699 |
+
"aggregation": "mean",
|
| 1700 |
+
"higher_is_better": true
|
| 1701 |
+
}
|
| 1702 |
+
],
|
| 1703 |
+
"output_type": "multiple_choice",
|
| 1704 |
+
"repeats": 1,
|
| 1705 |
+
"should_decontaminate": false,
|
| 1706 |
+
"metadata": {
|
| 1707 |
+
"version": 0.0
|
| 1708 |
+
}
|
| 1709 |
+
},
|
| 1710 |
+
"mmlu_logical_fallacies": {
|
| 1711 |
+
"task": "mmlu_logical_fallacies",
|
| 1712 |
+
"task_alias": "logical_fallacies",
|
| 1713 |
+
"group": "mmlu_humanities",
|
| 1714 |
+
"group_alias": "humanities",
|
| 1715 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1716 |
+
"dataset_name": "logical_fallacies",
|
| 1717 |
+
"test_split": "test",
|
| 1718 |
+
"fewshot_split": "dev",
|
| 1719 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1720 |
+
"doc_to_target": "answer",
|
| 1721 |
+
"doc_to_choice": [
|
| 1722 |
+
"A",
|
| 1723 |
+
"B",
|
| 1724 |
+
"C",
|
| 1725 |
+
"D"
|
| 1726 |
+
],
|
| 1727 |
+
"description": "The following are multiple choice questions (with answers) about logical fallacies.\n\n",
|
| 1728 |
+
"target_delimiter": " ",
|
| 1729 |
+
"fewshot_delimiter": "\n\n",
|
| 1730 |
+
"fewshot_config": {
|
| 1731 |
+
"sampler": "first_n"
|
| 1732 |
+
},
|
| 1733 |
+
"num_fewshot": 5,
|
| 1734 |
+
"metric_list": [
|
| 1735 |
+
{
|
| 1736 |
+
"metric": "acc",
|
| 1737 |
+
"aggregation": "mean",
|
| 1738 |
+
"higher_is_better": true
|
| 1739 |
+
}
|
| 1740 |
+
],
|
| 1741 |
+
"output_type": "multiple_choice",
|
| 1742 |
+
"repeats": 1,
|
| 1743 |
+
"should_decontaminate": false,
|
| 1744 |
+
"metadata": {
|
| 1745 |
+
"version": 0.0
|
| 1746 |
+
}
|
| 1747 |
+
},
|
| 1748 |
+
"mmlu_machine_learning": {
|
| 1749 |
+
"task": "mmlu_machine_learning",
|
| 1750 |
+
"task_alias": "machine_learning",
|
| 1751 |
+
"group": "mmlu_stem",
|
| 1752 |
+
"group_alias": "stem",
|
| 1753 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1754 |
+
"dataset_name": "machine_learning",
|
| 1755 |
+
"test_split": "test",
|
| 1756 |
+
"fewshot_split": "dev",
|
| 1757 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1758 |
+
"doc_to_target": "answer",
|
| 1759 |
+
"doc_to_choice": [
|
| 1760 |
+
"A",
|
| 1761 |
+
"B",
|
| 1762 |
+
"C",
|
| 1763 |
+
"D"
|
| 1764 |
+
],
|
| 1765 |
+
"description": "The following are multiple choice questions (with answers) about machine learning.\n\n",
|
| 1766 |
+
"target_delimiter": " ",
|
| 1767 |
+
"fewshot_delimiter": "\n\n",
|
| 1768 |
+
"fewshot_config": {
|
| 1769 |
+
"sampler": "first_n"
|
| 1770 |
+
},
|
| 1771 |
+
"num_fewshot": 5,
|
| 1772 |
+
"metric_list": [
|
| 1773 |
+
{
|
| 1774 |
+
"metric": "acc",
|
| 1775 |
+
"aggregation": "mean",
|
| 1776 |
+
"higher_is_better": true
|
| 1777 |
+
}
|
| 1778 |
+
],
|
| 1779 |
+
"output_type": "multiple_choice",
|
| 1780 |
+
"repeats": 1,
|
| 1781 |
+
"should_decontaminate": false,
|
| 1782 |
+
"metadata": {
|
| 1783 |
+
"version": 0.0
|
| 1784 |
+
}
|
| 1785 |
+
},
|
| 1786 |
+
"mmlu_management": {
|
| 1787 |
+
"task": "mmlu_management",
|
| 1788 |
+
"task_alias": "management",
|
| 1789 |
+
"group": "mmlu_other",
|
| 1790 |
+
"group_alias": "other",
|
| 1791 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1792 |
+
"dataset_name": "management",
|
| 1793 |
+
"test_split": "test",
|
| 1794 |
+
"fewshot_split": "dev",
|
| 1795 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1796 |
+
"doc_to_target": "answer",
|
| 1797 |
+
"doc_to_choice": [
|
| 1798 |
+
"A",
|
| 1799 |
+
"B",
|
| 1800 |
+
"C",
|
| 1801 |
+
"D"
|
| 1802 |
+
],
|
| 1803 |
+
"description": "The following are multiple choice questions (with answers) about management.\n\n",
|
| 1804 |
+
"target_delimiter": " ",
|
| 1805 |
+
"fewshot_delimiter": "\n\n",
|
| 1806 |
+
"fewshot_config": {
|
| 1807 |
+
"sampler": "first_n"
|
| 1808 |
+
},
|
| 1809 |
+
"num_fewshot": 5,
|
| 1810 |
+
"metric_list": [
|
| 1811 |
+
{
|
| 1812 |
+
"metric": "acc",
|
| 1813 |
+
"aggregation": "mean",
|
| 1814 |
+
"higher_is_better": true
|
| 1815 |
+
}
|
| 1816 |
+
],
|
| 1817 |
+
"output_type": "multiple_choice",
|
| 1818 |
+
"repeats": 1,
|
| 1819 |
+
"should_decontaminate": false,
|
| 1820 |
+
"metadata": {
|
| 1821 |
+
"version": 0.0
|
| 1822 |
+
}
|
| 1823 |
+
},
|
| 1824 |
+
"mmlu_marketing": {
|
| 1825 |
+
"task": "mmlu_marketing",
|
| 1826 |
+
"task_alias": "marketing",
|
| 1827 |
+
"group": "mmlu_other",
|
| 1828 |
+
"group_alias": "other",
|
| 1829 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1830 |
+
"dataset_name": "marketing",
|
| 1831 |
+
"test_split": "test",
|
| 1832 |
+
"fewshot_split": "dev",
|
| 1833 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1834 |
+
"doc_to_target": "answer",
|
| 1835 |
+
"doc_to_choice": [
|
| 1836 |
+
"A",
|
| 1837 |
+
"B",
|
| 1838 |
+
"C",
|
| 1839 |
+
"D"
|
| 1840 |
+
],
|
| 1841 |
+
"description": "The following are multiple choice questions (with answers) about marketing.\n\n",
|
| 1842 |
+
"target_delimiter": " ",
|
| 1843 |
+
"fewshot_delimiter": "\n\n",
|
| 1844 |
+
"fewshot_config": {
|
| 1845 |
+
"sampler": "first_n"
|
| 1846 |
+
},
|
| 1847 |
+
"num_fewshot": 5,
|
| 1848 |
+
"metric_list": [
|
| 1849 |
+
{
|
| 1850 |
+
"metric": "acc",
|
| 1851 |
+
"aggregation": "mean",
|
| 1852 |
+
"higher_is_better": true
|
| 1853 |
+
}
|
| 1854 |
+
],
|
| 1855 |
+
"output_type": "multiple_choice",
|
| 1856 |
+
"repeats": 1,
|
| 1857 |
+
"should_decontaminate": false,
|
| 1858 |
+
"metadata": {
|
| 1859 |
+
"version": 0.0
|
| 1860 |
+
}
|
| 1861 |
+
},
|
| 1862 |
+
"mmlu_medical_genetics": {
|
| 1863 |
+
"task": "mmlu_medical_genetics",
|
| 1864 |
+
"task_alias": "medical_genetics",
|
| 1865 |
+
"group": "mmlu_other",
|
| 1866 |
+
"group_alias": "other",
|
| 1867 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1868 |
+
"dataset_name": "medical_genetics",
|
| 1869 |
+
"test_split": "test",
|
| 1870 |
+
"fewshot_split": "dev",
|
| 1871 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1872 |
+
"doc_to_target": "answer",
|
| 1873 |
+
"doc_to_choice": [
|
| 1874 |
+
"A",
|
| 1875 |
+
"B",
|
| 1876 |
+
"C",
|
| 1877 |
+
"D"
|
| 1878 |
+
],
|
| 1879 |
+
"description": "The following are multiple choice questions (with answers) about medical genetics.\n\n",
|
| 1880 |
+
"target_delimiter": " ",
|
| 1881 |
+
"fewshot_delimiter": "\n\n",
|
| 1882 |
+
"fewshot_config": {
|
| 1883 |
+
"sampler": "first_n"
|
| 1884 |
+
},
|
| 1885 |
+
"num_fewshot": 5,
|
| 1886 |
+
"metric_list": [
|
| 1887 |
+
{
|
| 1888 |
+
"metric": "acc",
|
| 1889 |
+
"aggregation": "mean",
|
| 1890 |
+
"higher_is_better": true
|
| 1891 |
+
}
|
| 1892 |
+
],
|
| 1893 |
+
"output_type": "multiple_choice",
|
| 1894 |
+
"repeats": 1,
|
| 1895 |
+
"should_decontaminate": false,
|
| 1896 |
+
"metadata": {
|
| 1897 |
+
"version": 0.0
|
| 1898 |
+
}
|
| 1899 |
+
},
|
| 1900 |
+
"mmlu_miscellaneous": {
|
| 1901 |
+
"task": "mmlu_miscellaneous",
|
| 1902 |
+
"task_alias": "miscellaneous",
|
| 1903 |
+
"group": "mmlu_other",
|
| 1904 |
+
"group_alias": "other",
|
| 1905 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1906 |
+
"dataset_name": "miscellaneous",
|
| 1907 |
+
"test_split": "test",
|
| 1908 |
+
"fewshot_split": "dev",
|
| 1909 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1910 |
+
"doc_to_target": "answer",
|
| 1911 |
+
"doc_to_choice": [
|
| 1912 |
+
"A",
|
| 1913 |
+
"B",
|
| 1914 |
+
"C",
|
| 1915 |
+
"D"
|
| 1916 |
+
],
|
| 1917 |
+
"description": "The following are multiple choice questions (with answers) about miscellaneous.\n\n",
|
| 1918 |
+
"target_delimiter": " ",
|
| 1919 |
+
"fewshot_delimiter": "\n\n",
|
| 1920 |
+
"fewshot_config": {
|
| 1921 |
+
"sampler": "first_n"
|
| 1922 |
+
},
|
| 1923 |
+
"num_fewshot": 5,
|
| 1924 |
+
"metric_list": [
|
| 1925 |
+
{
|
| 1926 |
+
"metric": "acc",
|
| 1927 |
+
"aggregation": "mean",
|
| 1928 |
+
"higher_is_better": true
|
| 1929 |
+
}
|
| 1930 |
+
],
|
| 1931 |
+
"output_type": "multiple_choice",
|
| 1932 |
+
"repeats": 1,
|
| 1933 |
+
"should_decontaminate": false,
|
| 1934 |
+
"metadata": {
|
| 1935 |
+
"version": 0.0
|
| 1936 |
+
}
|
| 1937 |
+
},
|
| 1938 |
+
"mmlu_moral_disputes": {
|
| 1939 |
+
"task": "mmlu_moral_disputes",
|
| 1940 |
+
"task_alias": "moral_disputes",
|
| 1941 |
+
"group": "mmlu_humanities",
|
| 1942 |
+
"group_alias": "humanities",
|
| 1943 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1944 |
+
"dataset_name": "moral_disputes",
|
| 1945 |
+
"test_split": "test",
|
| 1946 |
+
"fewshot_split": "dev",
|
| 1947 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1948 |
+
"doc_to_target": "answer",
|
| 1949 |
+
"doc_to_choice": [
|
| 1950 |
+
"A",
|
| 1951 |
+
"B",
|
| 1952 |
+
"C",
|
| 1953 |
+
"D"
|
| 1954 |
+
],
|
| 1955 |
+
"description": "The following are multiple choice questions (with answers) about moral disputes.\n\n",
|
| 1956 |
+
"target_delimiter": " ",
|
| 1957 |
+
"fewshot_delimiter": "\n\n",
|
| 1958 |
+
"fewshot_config": {
|
| 1959 |
+
"sampler": "first_n"
|
| 1960 |
+
},
|
| 1961 |
+
"num_fewshot": 5,
|
| 1962 |
+
"metric_list": [
|
| 1963 |
+
{
|
| 1964 |
+
"metric": "acc",
|
| 1965 |
+
"aggregation": "mean",
|
| 1966 |
+
"higher_is_better": true
|
| 1967 |
+
}
|
| 1968 |
+
],
|
| 1969 |
+
"output_type": "multiple_choice",
|
| 1970 |
+
"repeats": 1,
|
| 1971 |
+
"should_decontaminate": false,
|
| 1972 |
+
"metadata": {
|
| 1973 |
+
"version": 0.0
|
| 1974 |
+
}
|
| 1975 |
+
},
|
| 1976 |
+
"mmlu_moral_scenarios": {
|
| 1977 |
+
"task": "mmlu_moral_scenarios",
|
| 1978 |
+
"task_alias": "moral_scenarios",
|
| 1979 |
+
"group": "mmlu_humanities",
|
| 1980 |
+
"group_alias": "humanities",
|
| 1981 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 1982 |
+
"dataset_name": "moral_scenarios",
|
| 1983 |
+
"test_split": "test",
|
| 1984 |
+
"fewshot_split": "dev",
|
| 1985 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1986 |
+
"doc_to_target": "answer",
|
| 1987 |
+
"doc_to_choice": [
|
| 1988 |
+
"A",
|
| 1989 |
+
"B",
|
| 1990 |
+
"C",
|
| 1991 |
+
"D"
|
| 1992 |
+
],
|
| 1993 |
+
"description": "The following are multiple choice questions (with answers) about moral scenarios.\n\n",
|
| 1994 |
+
"target_delimiter": " ",
|
| 1995 |
+
"fewshot_delimiter": "\n\n",
|
| 1996 |
+
"fewshot_config": {
|
| 1997 |
+
"sampler": "first_n"
|
| 1998 |
+
},
|
| 1999 |
+
"num_fewshot": 5,
|
| 2000 |
+
"metric_list": [
|
| 2001 |
+
{
|
| 2002 |
+
"metric": "acc",
|
| 2003 |
+
"aggregation": "mean",
|
| 2004 |
+
"higher_is_better": true
|
| 2005 |
+
}
|
| 2006 |
+
],
|
| 2007 |
+
"output_type": "multiple_choice",
|
| 2008 |
+
"repeats": 1,
|
| 2009 |
+
"should_decontaminate": false,
|
| 2010 |
+
"metadata": {
|
| 2011 |
+
"version": 0.0
|
| 2012 |
+
}
|
| 2013 |
+
},
|
| 2014 |
+
"mmlu_nutrition": {
|
| 2015 |
+
"task": "mmlu_nutrition",
|
| 2016 |
+
"task_alias": "nutrition",
|
| 2017 |
+
"group": "mmlu_other",
|
| 2018 |
+
"group_alias": "other",
|
| 2019 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2020 |
+
"dataset_name": "nutrition",
|
| 2021 |
+
"test_split": "test",
|
| 2022 |
+
"fewshot_split": "dev",
|
| 2023 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2024 |
+
"doc_to_target": "answer",
|
| 2025 |
+
"doc_to_choice": [
|
| 2026 |
+
"A",
|
| 2027 |
+
"B",
|
| 2028 |
+
"C",
|
| 2029 |
+
"D"
|
| 2030 |
+
],
|
| 2031 |
+
"description": "The following are multiple choice questions (with answers) about nutrition.\n\n",
|
| 2032 |
+
"target_delimiter": " ",
|
| 2033 |
+
"fewshot_delimiter": "\n\n",
|
| 2034 |
+
"fewshot_config": {
|
| 2035 |
+
"sampler": "first_n"
|
| 2036 |
+
},
|
| 2037 |
+
"num_fewshot": 5,
|
| 2038 |
+
"metric_list": [
|
| 2039 |
+
{
|
| 2040 |
+
"metric": "acc",
|
| 2041 |
+
"aggregation": "mean",
|
| 2042 |
+
"higher_is_better": true
|
| 2043 |
+
}
|
| 2044 |
+
],
|
| 2045 |
+
"output_type": "multiple_choice",
|
| 2046 |
+
"repeats": 1,
|
| 2047 |
+
"should_decontaminate": false,
|
| 2048 |
+
"metadata": {
|
| 2049 |
+
"version": 0.0
|
| 2050 |
+
}
|
| 2051 |
+
},
|
| 2052 |
+
"mmlu_philosophy": {
|
| 2053 |
+
"task": "mmlu_philosophy",
|
| 2054 |
+
"task_alias": "philosophy",
|
| 2055 |
+
"group": "mmlu_humanities",
|
| 2056 |
+
"group_alias": "humanities",
|
| 2057 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2058 |
+
"dataset_name": "philosophy",
|
| 2059 |
+
"test_split": "test",
|
| 2060 |
+
"fewshot_split": "dev",
|
| 2061 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2062 |
+
"doc_to_target": "answer",
|
| 2063 |
+
"doc_to_choice": [
|
| 2064 |
+
"A",
|
| 2065 |
+
"B",
|
| 2066 |
+
"C",
|
| 2067 |
+
"D"
|
| 2068 |
+
],
|
| 2069 |
+
"description": "The following are multiple choice questions (with answers) about philosophy.\n\n",
|
| 2070 |
+
"target_delimiter": " ",
|
| 2071 |
+
"fewshot_delimiter": "\n\n",
|
| 2072 |
+
"fewshot_config": {
|
| 2073 |
+
"sampler": "first_n"
|
| 2074 |
+
},
|
| 2075 |
+
"num_fewshot": 5,
|
| 2076 |
+
"metric_list": [
|
| 2077 |
+
{
|
| 2078 |
+
"metric": "acc",
|
| 2079 |
+
"aggregation": "mean",
|
| 2080 |
+
"higher_is_better": true
|
| 2081 |
+
}
|
| 2082 |
+
],
|
| 2083 |
+
"output_type": "multiple_choice",
|
| 2084 |
+
"repeats": 1,
|
| 2085 |
+
"should_decontaminate": false,
|
| 2086 |
+
"metadata": {
|
| 2087 |
+
"version": 0.0
|
| 2088 |
+
}
|
| 2089 |
+
},
|
| 2090 |
+
"mmlu_prehistory": {
|
| 2091 |
+
"task": "mmlu_prehistory",
|
| 2092 |
+
"task_alias": "prehistory",
|
| 2093 |
+
"group": "mmlu_humanities",
|
| 2094 |
+
"group_alias": "humanities",
|
| 2095 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2096 |
+
"dataset_name": "prehistory",
|
| 2097 |
+
"test_split": "test",
|
| 2098 |
+
"fewshot_split": "dev",
|
| 2099 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2100 |
+
"doc_to_target": "answer",
|
| 2101 |
+
"doc_to_choice": [
|
| 2102 |
+
"A",
|
| 2103 |
+
"B",
|
| 2104 |
+
"C",
|
| 2105 |
+
"D"
|
| 2106 |
+
],
|
| 2107 |
+
"description": "The following are multiple choice questions (with answers) about prehistory.\n\n",
|
| 2108 |
+
"target_delimiter": " ",
|
| 2109 |
+
"fewshot_delimiter": "\n\n",
|
| 2110 |
+
"fewshot_config": {
|
| 2111 |
+
"sampler": "first_n"
|
| 2112 |
+
},
|
| 2113 |
+
"num_fewshot": 5,
|
| 2114 |
+
"metric_list": [
|
| 2115 |
+
{
|
| 2116 |
+
"metric": "acc",
|
| 2117 |
+
"aggregation": "mean",
|
| 2118 |
+
"higher_is_better": true
|
| 2119 |
+
}
|
| 2120 |
+
],
|
| 2121 |
+
"output_type": "multiple_choice",
|
| 2122 |
+
"repeats": 1,
|
| 2123 |
+
"should_decontaminate": false,
|
| 2124 |
+
"metadata": {
|
| 2125 |
+
"version": 0.0
|
| 2126 |
+
}
|
| 2127 |
+
},
|
| 2128 |
+
"mmlu_professional_accounting": {
|
| 2129 |
+
"task": "mmlu_professional_accounting",
|
| 2130 |
+
"task_alias": "professional_accounting",
|
| 2131 |
+
"group": "mmlu_other",
|
| 2132 |
+
"group_alias": "other",
|
| 2133 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2134 |
+
"dataset_name": "professional_accounting",
|
| 2135 |
+
"test_split": "test",
|
| 2136 |
+
"fewshot_split": "dev",
|
| 2137 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2138 |
+
"doc_to_target": "answer",
|
| 2139 |
+
"doc_to_choice": [
|
| 2140 |
+
"A",
|
| 2141 |
+
"B",
|
| 2142 |
+
"C",
|
| 2143 |
+
"D"
|
| 2144 |
+
],
|
| 2145 |
+
"description": "The following are multiple choice questions (with answers) about professional accounting.\n\n",
|
| 2146 |
+
"target_delimiter": " ",
|
| 2147 |
+
"fewshot_delimiter": "\n\n",
|
| 2148 |
+
"fewshot_config": {
|
| 2149 |
+
"sampler": "first_n"
|
| 2150 |
+
},
|
| 2151 |
+
"num_fewshot": 5,
|
| 2152 |
+
"metric_list": [
|
| 2153 |
+
{
|
| 2154 |
+
"metric": "acc",
|
| 2155 |
+
"aggregation": "mean",
|
| 2156 |
+
"higher_is_better": true
|
| 2157 |
+
}
|
| 2158 |
+
],
|
| 2159 |
+
"output_type": "multiple_choice",
|
| 2160 |
+
"repeats": 1,
|
| 2161 |
+
"should_decontaminate": false,
|
| 2162 |
+
"metadata": {
|
| 2163 |
+
"version": 0.0
|
| 2164 |
+
}
|
| 2165 |
+
},
|
| 2166 |
+
"mmlu_professional_law": {
|
| 2167 |
+
"task": "mmlu_professional_law",
|
| 2168 |
+
"task_alias": "professional_law",
|
| 2169 |
+
"group": "mmlu_humanities",
|
| 2170 |
+
"group_alias": "humanities",
|
| 2171 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2172 |
+
"dataset_name": "professional_law",
|
| 2173 |
+
"test_split": "test",
|
| 2174 |
+
"fewshot_split": "dev",
|
| 2175 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2176 |
+
"doc_to_target": "answer",
|
| 2177 |
+
"doc_to_choice": [
|
| 2178 |
+
"A",
|
| 2179 |
+
"B",
|
| 2180 |
+
"C",
|
| 2181 |
+
"D"
|
| 2182 |
+
],
|
| 2183 |
+
"description": "The following are multiple choice questions (with answers) about professional law.\n\n",
|
| 2184 |
+
"target_delimiter": " ",
|
| 2185 |
+
"fewshot_delimiter": "\n\n",
|
| 2186 |
+
"fewshot_config": {
|
| 2187 |
+
"sampler": "first_n"
|
| 2188 |
+
},
|
| 2189 |
+
"num_fewshot": 5,
|
| 2190 |
+
"metric_list": [
|
| 2191 |
+
{
|
| 2192 |
+
"metric": "acc",
|
| 2193 |
+
"aggregation": "mean",
|
| 2194 |
+
"higher_is_better": true
|
| 2195 |
+
}
|
| 2196 |
+
],
|
| 2197 |
+
"output_type": "multiple_choice",
|
| 2198 |
+
"repeats": 1,
|
| 2199 |
+
"should_decontaminate": false,
|
| 2200 |
+
"metadata": {
|
| 2201 |
+
"version": 0.0
|
| 2202 |
+
}
|
| 2203 |
+
},
|
| 2204 |
+
"mmlu_professional_medicine": {
|
| 2205 |
+
"task": "mmlu_professional_medicine",
|
| 2206 |
+
"task_alias": "professional_medicine",
|
| 2207 |
+
"group": "mmlu_other",
|
| 2208 |
+
"group_alias": "other",
|
| 2209 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2210 |
+
"dataset_name": "professional_medicine",
|
| 2211 |
+
"test_split": "test",
|
| 2212 |
+
"fewshot_split": "dev",
|
| 2213 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2214 |
+
"doc_to_target": "answer",
|
| 2215 |
+
"doc_to_choice": [
|
| 2216 |
+
"A",
|
| 2217 |
+
"B",
|
| 2218 |
+
"C",
|
| 2219 |
+
"D"
|
| 2220 |
+
],
|
| 2221 |
+
"description": "The following are multiple choice questions (with answers) about professional medicine.\n\n",
|
| 2222 |
+
"target_delimiter": " ",
|
| 2223 |
+
"fewshot_delimiter": "\n\n",
|
| 2224 |
+
"fewshot_config": {
|
| 2225 |
+
"sampler": "first_n"
|
| 2226 |
+
},
|
| 2227 |
+
"num_fewshot": 5,
|
| 2228 |
+
"metric_list": [
|
| 2229 |
+
{
|
| 2230 |
+
"metric": "acc",
|
| 2231 |
+
"aggregation": "mean",
|
| 2232 |
+
"higher_is_better": true
|
| 2233 |
+
}
|
| 2234 |
+
],
|
| 2235 |
+
"output_type": "multiple_choice",
|
| 2236 |
+
"repeats": 1,
|
| 2237 |
+
"should_decontaminate": false,
|
| 2238 |
+
"metadata": {
|
| 2239 |
+
"version": 0.0
|
| 2240 |
+
}
|
| 2241 |
+
},
|
| 2242 |
+
"mmlu_professional_psychology": {
|
| 2243 |
+
"task": "mmlu_professional_psychology",
|
| 2244 |
+
"task_alias": "professional_psychology",
|
| 2245 |
+
"group": "mmlu_social_sciences",
|
| 2246 |
+
"group_alias": "social_sciences",
|
| 2247 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2248 |
+
"dataset_name": "professional_psychology",
|
| 2249 |
+
"test_split": "test",
|
| 2250 |
+
"fewshot_split": "dev",
|
| 2251 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2252 |
+
"doc_to_target": "answer",
|
| 2253 |
+
"doc_to_choice": [
|
| 2254 |
+
"A",
|
| 2255 |
+
"B",
|
| 2256 |
+
"C",
|
| 2257 |
+
"D"
|
| 2258 |
+
],
|
| 2259 |
+
"description": "The following are multiple choice questions (with answers) about professional psychology.\n\n",
|
| 2260 |
+
"target_delimiter": " ",
|
| 2261 |
+
"fewshot_delimiter": "\n\n",
|
| 2262 |
+
"fewshot_config": {
|
| 2263 |
+
"sampler": "first_n"
|
| 2264 |
+
},
|
| 2265 |
+
"num_fewshot": 5,
|
| 2266 |
+
"metric_list": [
|
| 2267 |
+
{
|
| 2268 |
+
"metric": "acc",
|
| 2269 |
+
"aggregation": "mean",
|
| 2270 |
+
"higher_is_better": true
|
| 2271 |
+
}
|
| 2272 |
+
],
|
| 2273 |
+
"output_type": "multiple_choice",
|
| 2274 |
+
"repeats": 1,
|
| 2275 |
+
"should_decontaminate": false,
|
| 2276 |
+
"metadata": {
|
| 2277 |
+
"version": 0.0
|
| 2278 |
+
}
|
| 2279 |
+
},
|
| 2280 |
+
"mmlu_public_relations": {
|
| 2281 |
+
"task": "mmlu_public_relations",
|
| 2282 |
+
"task_alias": "public_relations",
|
| 2283 |
+
"group": "mmlu_social_sciences",
|
| 2284 |
+
"group_alias": "social_sciences",
|
| 2285 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2286 |
+
"dataset_name": "public_relations",
|
| 2287 |
+
"test_split": "test",
|
| 2288 |
+
"fewshot_split": "dev",
|
| 2289 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2290 |
+
"doc_to_target": "answer",
|
| 2291 |
+
"doc_to_choice": [
|
| 2292 |
+
"A",
|
| 2293 |
+
"B",
|
| 2294 |
+
"C",
|
| 2295 |
+
"D"
|
| 2296 |
+
],
|
| 2297 |
+
"description": "The following are multiple choice questions (with answers) about public relations.\n\n",
|
| 2298 |
+
"target_delimiter": " ",
|
| 2299 |
+
"fewshot_delimiter": "\n\n",
|
| 2300 |
+
"fewshot_config": {
|
| 2301 |
+
"sampler": "first_n"
|
| 2302 |
+
},
|
| 2303 |
+
"num_fewshot": 5,
|
| 2304 |
+
"metric_list": [
|
| 2305 |
+
{
|
| 2306 |
+
"metric": "acc",
|
| 2307 |
+
"aggregation": "mean",
|
| 2308 |
+
"higher_is_better": true
|
| 2309 |
+
}
|
| 2310 |
+
],
|
| 2311 |
+
"output_type": "multiple_choice",
|
| 2312 |
+
"repeats": 1,
|
| 2313 |
+
"should_decontaminate": false,
|
| 2314 |
+
"metadata": {
|
| 2315 |
+
"version": 0.0
|
| 2316 |
+
}
|
| 2317 |
+
},
|
| 2318 |
+
"mmlu_security_studies": {
|
| 2319 |
+
"task": "mmlu_security_studies",
|
| 2320 |
+
"task_alias": "security_studies",
|
| 2321 |
+
"group": "mmlu_social_sciences",
|
| 2322 |
+
"group_alias": "social_sciences",
|
| 2323 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2324 |
+
"dataset_name": "security_studies",
|
| 2325 |
+
"test_split": "test",
|
| 2326 |
+
"fewshot_split": "dev",
|
| 2327 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2328 |
+
"doc_to_target": "answer",
|
| 2329 |
+
"doc_to_choice": [
|
| 2330 |
+
"A",
|
| 2331 |
+
"B",
|
| 2332 |
+
"C",
|
| 2333 |
+
"D"
|
| 2334 |
+
],
|
| 2335 |
+
"description": "The following are multiple choice questions (with answers) about security studies.\n\n",
|
| 2336 |
+
"target_delimiter": " ",
|
| 2337 |
+
"fewshot_delimiter": "\n\n",
|
| 2338 |
+
"fewshot_config": {
|
| 2339 |
+
"sampler": "first_n"
|
| 2340 |
+
},
|
| 2341 |
+
"num_fewshot": 5,
|
| 2342 |
+
"metric_list": [
|
| 2343 |
+
{
|
| 2344 |
+
"metric": "acc",
|
| 2345 |
+
"aggregation": "mean",
|
| 2346 |
+
"higher_is_better": true
|
| 2347 |
+
}
|
| 2348 |
+
],
|
| 2349 |
+
"output_type": "multiple_choice",
|
| 2350 |
+
"repeats": 1,
|
| 2351 |
+
"should_decontaminate": false,
|
| 2352 |
+
"metadata": {
|
| 2353 |
+
"version": 0.0
|
| 2354 |
+
}
|
| 2355 |
+
},
|
| 2356 |
+
"mmlu_sociology": {
|
| 2357 |
+
"task": "mmlu_sociology",
|
| 2358 |
+
"task_alias": "sociology",
|
| 2359 |
+
"group": "mmlu_social_sciences",
|
| 2360 |
+
"group_alias": "social_sciences",
|
| 2361 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2362 |
+
"dataset_name": "sociology",
|
| 2363 |
+
"test_split": "test",
|
| 2364 |
+
"fewshot_split": "dev",
|
| 2365 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2366 |
+
"doc_to_target": "answer",
|
| 2367 |
+
"doc_to_choice": [
|
| 2368 |
+
"A",
|
| 2369 |
+
"B",
|
| 2370 |
+
"C",
|
| 2371 |
+
"D"
|
| 2372 |
+
],
|
| 2373 |
+
"description": "The following are multiple choice questions (with answers) about sociology.\n\n",
|
| 2374 |
+
"target_delimiter": " ",
|
| 2375 |
+
"fewshot_delimiter": "\n\n",
|
| 2376 |
+
"fewshot_config": {
|
| 2377 |
+
"sampler": "first_n"
|
| 2378 |
+
},
|
| 2379 |
+
"num_fewshot": 5,
|
| 2380 |
+
"metric_list": [
|
| 2381 |
+
{
|
| 2382 |
+
"metric": "acc",
|
| 2383 |
+
"aggregation": "mean",
|
| 2384 |
+
"higher_is_better": true
|
| 2385 |
+
}
|
| 2386 |
+
],
|
| 2387 |
+
"output_type": "multiple_choice",
|
| 2388 |
+
"repeats": 1,
|
| 2389 |
+
"should_decontaminate": false,
|
| 2390 |
+
"metadata": {
|
| 2391 |
+
"version": 0.0
|
| 2392 |
+
}
|
| 2393 |
+
},
|
| 2394 |
+
"mmlu_us_foreign_policy": {
|
| 2395 |
+
"task": "mmlu_us_foreign_policy",
|
| 2396 |
+
"task_alias": "us_foreign_policy",
|
| 2397 |
+
"group": "mmlu_social_sciences",
|
| 2398 |
+
"group_alias": "social_sciences",
|
| 2399 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2400 |
+
"dataset_name": "us_foreign_policy",
|
| 2401 |
+
"test_split": "test",
|
| 2402 |
+
"fewshot_split": "dev",
|
| 2403 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2404 |
+
"doc_to_target": "answer",
|
| 2405 |
+
"doc_to_choice": [
|
| 2406 |
+
"A",
|
| 2407 |
+
"B",
|
| 2408 |
+
"C",
|
| 2409 |
+
"D"
|
| 2410 |
+
],
|
| 2411 |
+
"description": "The following are multiple choice questions (with answers) about us foreign policy.\n\n",
|
| 2412 |
+
"target_delimiter": " ",
|
| 2413 |
+
"fewshot_delimiter": "\n\n",
|
| 2414 |
+
"fewshot_config": {
|
| 2415 |
+
"sampler": "first_n"
|
| 2416 |
+
},
|
| 2417 |
+
"num_fewshot": 5,
|
| 2418 |
+
"metric_list": [
|
| 2419 |
+
{
|
| 2420 |
+
"metric": "acc",
|
| 2421 |
+
"aggregation": "mean",
|
| 2422 |
+
"higher_is_better": true
|
| 2423 |
+
}
|
| 2424 |
+
],
|
| 2425 |
+
"output_type": "multiple_choice",
|
| 2426 |
+
"repeats": 1,
|
| 2427 |
+
"should_decontaminate": false,
|
| 2428 |
+
"metadata": {
|
| 2429 |
+
"version": 0.0
|
| 2430 |
+
}
|
| 2431 |
+
},
|
| 2432 |
+
"mmlu_virology": {
|
| 2433 |
+
"task": "mmlu_virology",
|
| 2434 |
+
"task_alias": "virology",
|
| 2435 |
+
"group": "mmlu_other",
|
| 2436 |
+
"group_alias": "other",
|
| 2437 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2438 |
+
"dataset_name": "virology",
|
| 2439 |
+
"test_split": "test",
|
| 2440 |
+
"fewshot_split": "dev",
|
| 2441 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2442 |
+
"doc_to_target": "answer",
|
| 2443 |
+
"doc_to_choice": [
|
| 2444 |
+
"A",
|
| 2445 |
+
"B",
|
| 2446 |
+
"C",
|
| 2447 |
+
"D"
|
| 2448 |
+
],
|
| 2449 |
+
"description": "The following are multiple choice questions (with answers) about virology.\n\n",
|
| 2450 |
+
"target_delimiter": " ",
|
| 2451 |
+
"fewshot_delimiter": "\n\n",
|
| 2452 |
+
"fewshot_config": {
|
| 2453 |
+
"sampler": "first_n"
|
| 2454 |
+
},
|
| 2455 |
+
"num_fewshot": 5,
|
| 2456 |
+
"metric_list": [
|
| 2457 |
+
{
|
| 2458 |
+
"metric": "acc",
|
| 2459 |
+
"aggregation": "mean",
|
| 2460 |
+
"higher_is_better": true
|
| 2461 |
+
}
|
| 2462 |
+
],
|
| 2463 |
+
"output_type": "multiple_choice",
|
| 2464 |
+
"repeats": 1,
|
| 2465 |
+
"should_decontaminate": false,
|
| 2466 |
+
"metadata": {
|
| 2467 |
+
"version": 0.0
|
| 2468 |
+
}
|
| 2469 |
+
},
|
| 2470 |
+
"mmlu_world_religions": {
|
| 2471 |
+
"task": "mmlu_world_religions",
|
| 2472 |
+
"task_alias": "world_religions",
|
| 2473 |
+
"group": "mmlu_humanities",
|
| 2474 |
+
"group_alias": "humanities",
|
| 2475 |
+
"dataset_path": "hails/mmlu_no_train",
|
| 2476 |
+
"dataset_name": "world_religions",
|
| 2477 |
+
"test_split": "test",
|
| 2478 |
+
"fewshot_split": "dev",
|
| 2479 |
+
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2480 |
+
"doc_to_target": "answer",
|
| 2481 |
+
"doc_to_choice": [
|
| 2482 |
+
"A",
|
| 2483 |
+
"B",
|
| 2484 |
+
"C",
|
| 2485 |
+
"D"
|
| 2486 |
+
],
|
| 2487 |
+
"description": "The following are multiple choice questions (with answers) about world religions.\n\n",
|
| 2488 |
+
"target_delimiter": " ",
|
| 2489 |
+
"fewshot_delimiter": "\n\n",
|
| 2490 |
+
"fewshot_config": {
|
| 2491 |
+
"sampler": "first_n"
|
| 2492 |
+
},
|
| 2493 |
+
"num_fewshot": 5,
|
| 2494 |
+
"metric_list": [
|
| 2495 |
+
{
|
| 2496 |
+
"metric": "acc",
|
| 2497 |
+
"aggregation": "mean",
|
| 2498 |
+
"higher_is_better": true
|
| 2499 |
+
}
|
| 2500 |
+
],
|
| 2501 |
+
"output_type": "multiple_choice",
|
| 2502 |
+
"repeats": 1,
|
| 2503 |
+
"should_decontaminate": false,
|
| 2504 |
+
"metadata": {
|
| 2505 |
+
"version": 0.0
|
| 2506 |
+
}
|
| 2507 |
+
}
|
| 2508 |
+
},
|
| 2509 |
+
"versions": {
|
| 2510 |
+
"mmlu": "N/A",
|
| 2511 |
+
"mmlu_abstract_algebra": 0.0,
|
| 2512 |
+
"mmlu_anatomy": 0.0,
|
| 2513 |
+
"mmlu_astronomy": 0.0,
|
| 2514 |
+
"mmlu_business_ethics": 0.0,
|
| 2515 |
+
"mmlu_clinical_knowledge": 0.0,
|
| 2516 |
+
"mmlu_college_biology": 0.0,
|
| 2517 |
+
"mmlu_college_chemistry": 0.0,
|
| 2518 |
+
"mmlu_college_computer_science": 0.0,
|
| 2519 |
+
"mmlu_college_mathematics": 0.0,
|
| 2520 |
+
"mmlu_college_medicine": 0.0,
|
| 2521 |
+
"mmlu_college_physics": 0.0,
|
| 2522 |
+
"mmlu_computer_security": 0.0,
|
| 2523 |
+
"mmlu_conceptual_physics": 0.0,
|
| 2524 |
+
"mmlu_econometrics": 0.0,
|
| 2525 |
+
"mmlu_electrical_engineering": 0.0,
|
| 2526 |
+
"mmlu_elementary_mathematics": 0.0,
|
| 2527 |
+
"mmlu_formal_logic": 0.0,
|
| 2528 |
+
"mmlu_global_facts": 0.0,
|
| 2529 |
+
"mmlu_high_school_biology": 0.0,
|
| 2530 |
+
"mmlu_high_school_chemistry": 0.0,
|
| 2531 |
+
"mmlu_high_school_computer_science": 0.0,
|
| 2532 |
+
"mmlu_high_school_european_history": 0.0,
|
| 2533 |
+
"mmlu_high_school_geography": 0.0,
|
| 2534 |
+
"mmlu_high_school_government_and_politics": 0.0,
|
| 2535 |
+
"mmlu_high_school_macroeconomics": 0.0,
|
| 2536 |
+
"mmlu_high_school_mathematics": 0.0,
|
| 2537 |
+
"mmlu_high_school_microeconomics": 0.0,
|
| 2538 |
+
"mmlu_high_school_physics": 0.0,
|
| 2539 |
+
"mmlu_high_school_psychology": 0.0,
|
| 2540 |
+
"mmlu_high_school_statistics": 0.0,
|
| 2541 |
+
"mmlu_high_school_us_history": 0.0,
|
| 2542 |
+
"mmlu_high_school_world_history": 0.0,
|
| 2543 |
+
"mmlu_human_aging": 0.0,
|
| 2544 |
+
"mmlu_human_sexuality": 0.0,
|
| 2545 |
+
"mmlu_humanities": "N/A",
|
| 2546 |
+
"mmlu_international_law": 0.0,
|
| 2547 |
+
"mmlu_jurisprudence": 0.0,
|
| 2548 |
+
"mmlu_logical_fallacies": 0.0,
|
| 2549 |
+
"mmlu_machine_learning": 0.0,
|
| 2550 |
+
"mmlu_management": 0.0,
|
| 2551 |
+
"mmlu_marketing": 0.0,
|
| 2552 |
+
"mmlu_medical_genetics": 0.0,
|
| 2553 |
+
"mmlu_miscellaneous": 0.0,
|
| 2554 |
+
"mmlu_moral_disputes": 0.0,
|
| 2555 |
+
"mmlu_moral_scenarios": 0.0,
|
| 2556 |
+
"mmlu_nutrition": 0.0,
|
| 2557 |
+
"mmlu_other": "N/A",
|
| 2558 |
+
"mmlu_philosophy": 0.0,
|
| 2559 |
+
"mmlu_prehistory": 0.0,
|
| 2560 |
+
"mmlu_professional_accounting": 0.0,
|
| 2561 |
+
"mmlu_professional_law": 0.0,
|
| 2562 |
+
"mmlu_professional_medicine": 0.0,
|
| 2563 |
+
"mmlu_professional_psychology": 0.0,
|
| 2564 |
+
"mmlu_public_relations": 0.0,
|
| 2565 |
+
"mmlu_security_studies": 0.0,
|
| 2566 |
+
"mmlu_social_sciences": "N/A",
|
| 2567 |
+
"mmlu_sociology": 0.0,
|
| 2568 |
+
"mmlu_stem": "N/A",
|
| 2569 |
+
"mmlu_us_foreign_policy": 0.0,
|
| 2570 |
+
"mmlu_virology": 0.0,
|
| 2571 |
+
"mmlu_world_religions": 0.0
|
| 2572 |
+
},
|
| 2573 |
+
"n-shot": {
|
| 2574 |
+
"mmlu": 0,
|
| 2575 |
+
"mmlu_abstract_algebra": 5,
|
| 2576 |
+
"mmlu_anatomy": 5,
|
| 2577 |
+
"mmlu_astronomy": 5,
|
| 2578 |
+
"mmlu_business_ethics": 5,
|
| 2579 |
+
"mmlu_clinical_knowledge": 5,
|
| 2580 |
+
"mmlu_college_biology": 5,
|
| 2581 |
+
"mmlu_college_chemistry": 5,
|
| 2582 |
+
"mmlu_college_computer_science": 5,
|
| 2583 |
+
"mmlu_college_mathematics": 5,
|
| 2584 |
+
"mmlu_college_medicine": 5,
|
| 2585 |
+
"mmlu_college_physics": 5,
|
| 2586 |
+
"mmlu_computer_security": 5,
|
| 2587 |
+
"mmlu_conceptual_physics": 5,
|
| 2588 |
+
"mmlu_econometrics": 5,
|
| 2589 |
+
"mmlu_electrical_engineering": 5,
|
| 2590 |
+
"mmlu_elementary_mathematics": 5,
|
| 2591 |
+
"mmlu_formal_logic": 5,
|
| 2592 |
+
"mmlu_global_facts": 5,
|
| 2593 |
+
"mmlu_high_school_biology": 5,
|
| 2594 |
+
"mmlu_high_school_chemistry": 5,
|
| 2595 |
+
"mmlu_high_school_computer_science": 5,
|
| 2596 |
+
"mmlu_high_school_european_history": 5,
|
| 2597 |
+
"mmlu_high_school_geography": 5,
|
| 2598 |
+
"mmlu_high_school_government_and_politics": 5,
|
| 2599 |
+
"mmlu_high_school_macroeconomics": 5,
|
| 2600 |
+
"mmlu_high_school_mathematics": 5,
|
| 2601 |
+
"mmlu_high_school_microeconomics": 5,
|
| 2602 |
+
"mmlu_high_school_physics": 5,
|
| 2603 |
+
"mmlu_high_school_psychology": 5,
|
| 2604 |
+
"mmlu_high_school_statistics": 5,
|
| 2605 |
+
"mmlu_high_school_us_history": 5,
|
| 2606 |
+
"mmlu_high_school_world_history": 5,
|
| 2607 |
+
"mmlu_human_aging": 5,
|
| 2608 |
+
"mmlu_human_sexuality": 5,
|
| 2609 |
+
"mmlu_humanities": 5,
|
| 2610 |
+
"mmlu_international_law": 5,
|
| 2611 |
+
"mmlu_jurisprudence": 5,
|
| 2612 |
+
"mmlu_logical_fallacies": 5,
|
| 2613 |
+
"mmlu_machine_learning": 5,
|
| 2614 |
+
"mmlu_management": 5,
|
| 2615 |
+
"mmlu_marketing": 5,
|
| 2616 |
+
"mmlu_medical_genetics": 5,
|
| 2617 |
+
"mmlu_miscellaneous": 5,
|
| 2618 |
+
"mmlu_moral_disputes": 5,
|
| 2619 |
+
"mmlu_moral_scenarios": 5,
|
| 2620 |
+
"mmlu_nutrition": 5,
|
| 2621 |
+
"mmlu_other": 5,
|
| 2622 |
+
"mmlu_philosophy": 5,
|
| 2623 |
+
"mmlu_prehistory": 5,
|
| 2624 |
+
"mmlu_professional_accounting": 5,
|
| 2625 |
+
"mmlu_professional_law": 5,
|
| 2626 |
+
"mmlu_professional_medicine": 5,
|
| 2627 |
+
"mmlu_professional_psychology": 5,
|
| 2628 |
+
"mmlu_public_relations": 5,
|
| 2629 |
+
"mmlu_security_studies": 5,
|
| 2630 |
+
"mmlu_social_sciences": 5,
|
| 2631 |
+
"mmlu_sociology": 5,
|
| 2632 |
+
"mmlu_stem": 5,
|
| 2633 |
+
"mmlu_us_foreign_policy": 5,
|
| 2634 |
+
"mmlu_virology": 5,
|
| 2635 |
+
"mmlu_world_religions": 5
|
| 2636 |
+
},
|
| 2637 |
+
"config": {
|
| 2638 |
+
"model": "hf",
|
| 2639 |
+
"model_args": "pretrained=HuggingFaceH4/mistral-7b-ift,revision=v31.3,dtype=bfloat16",
|
| 2640 |
+
"batch_size": "auto",
|
| 2641 |
+
"batch_sizes": [
|
| 2642 |
+
16
|
| 2643 |
+
],
|
| 2644 |
+
"device": null,
|
| 2645 |
+
"use_cache": null,
|
| 2646 |
+
"limit": null,
|
| 2647 |
+
"bootstrap_iters": 100000,
|
| 2648 |
+
"gen_kwargs": null
|
| 2649 |
+
},
|
| 2650 |
+
"git_hash": "901c20b"
|
| 2651 |
+
}
|