WNJXYK/

LawBench_Results

Below is the format of the inference results in this repository. { "origin_prompt": "Question", "prediction": "Model Prediction", "refr": "Answer", "reasoning": "The Reasoning of the Prediction (Only available for R-Series Models)" }

	Performance of Deepseek-Series Models on LawBench




	

Deepseek-R1(LLAMA-8B) Deepseek-R1(LLAMA-70B) Deepseek-V3 Deepseek-R1

1-1 12.35 13.92 36.33 50.32

1-2 34.6 53 75.8 85.8… See the full description on the dataset page: https://huggingface.co/datasets/WNJXYK/LawBench_Results.