AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

CMedBench: A Comprehensive Benchmark for Efficient Medical Large Language Models

Shengbo Gao, Jinyang Guo, Lixian Su, Yifu Ding, Shiqiao Gu, Aishan Liu, Yuqing Ma, Zhiwang Zhang, Xianglong Liu

AAAI 2026 · Large language models › Language, safety and interpretability · approach: LLM pruning and scaling data · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply 4-bit weight-only quantization (AWQ/GPTQ)medical LLMs · medical knowledge tasksCPS: CPS 0.99 (AWQ), 0.97 (GPTQ)No clear effect
apply 4-bit SmoothQuant weight-activation quantizationmedical LLMs · medical knowledge tasksCPS: CPS 0.80Hurt
apply Wanda 50% unstructured sparsitymedical LLMs · medical knowledge tasksCPS: CPS 0.90Hurt
apply ShortGPT 25% layer pruningmedical LLMs · medical knowledge tasksCPS: CPS 0.74Hurt
apply Wanda 50% sparsitymedical LLMs · medical application tasksCPS: CPS 0.96No clear effect
apply sparsity on Jmed datasetmedical LLMs · JmedaccuracyHurt
scale up model size under ShortGPTLLaMA3, Qwen2.5, HuatuoGPT-o1, Meditron · MMLU-HealthCCS: Qwen2.5-7B 0.87 -> Qwen2.5-72B 0.96Improved
apply 8-bit quantization instead of 4-bitGPTQ/AWQ/SmoothQuant · MMLU-Healthaverage normalized ratio: SmoothQuant 0.77 -> 1.00Improved
apply Wanda sparsity 25% to 75%medical LLMs · MMLU-Healthaverage normalized ratio: 0.99 -> 0.69Hurt
apply quantizationmedical LLMs · trustworthiness (Fairness)TWYMixed
apply unstructured sparsity WandavLLM inference · efficiencyspeed and memoryNo clear effect
apply quantizationmedical LLMs · efficiencymemory and token rateImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy