ChoiceBench is a lightweight framework for MCQ evaluation-method research on LLMs, with built-in support for answer-order bias analysis and mitigation methods.
Proper scoring rules, reduces LLM overconfidence in multiple-choice QA.
中文问答能力评测：四家厂商五个模型 × 准确率/稳定性/安全，附判分口径、格式约束、思考开关与集间配比的单因子受控实验；全部数字可离线复算。
Open LLM leaderboard featuring Xiaomi MiMo v2.5 & MiMo 100T head-to-head with GPT-5, Claude, Gemini, DeepSeek, Llama 4. ARC-AGI · SWE-Bench · MMLU-Pro · GPQA · HumanEval · BFCL.
Self-hosted LLM evaluation lab: MMLU-Pro / BFCL / IFEval / LiveCodeBench / HumanEval+ runs on Tesla V100 and RTX 3090, llama.cpp serving (incl. Volta sm_70), patched API benchmark clients
Qwen3-32B quantization results on H100: speed, memory and accuracy. Documentation summary of AIMultiple research.
A Descriptive Analysis of Capacity, Training Stage, and Performance in Open Weight LLMs
Artifacts for confidence-stratified human and LLM-assisted auditing on MMLU-Pro and MMMU-Pro.
To associate your repository with the mmlu-pro topic, visit your repo's landing page and select "manage topics."