Andyyyy64/whichllm — 本地 LLM 硬件匹配 CLI(3.9k Stars)
Find the best local LLM that actually runs on your hardware. Auto-detects GPU/CPU/RAM, ranks top models from HuggingFace that fit — scored by real benchmarks, not parameter count.
核心痛点解决
"我的破显卡到底能跑什么模型?" — 硬件能力探测 + 本地 LLM 选型评估。
核心功能
- Auto-detect hardware — NVIDIA / AMD / Apple Silicon / CPU-only
- Evidence-based ranking — 合并 LiveBench / Artificial Analysis / Aider / Vision / Chatbot Arena ELO / Open LLM Leaderboard,按置信度加权
- Recency-aware — 沿模型家族线衰减旧评分,防止过时数据长期占优
- VRAM estimation — weights + KV cache + activation + overhead(~500MB)
- Speed modeling — 基于带宽/量化效率/后端/MoE 激活参数比估算 tok/s
- GPU simulation —
whichllm --gpu "RTX 4090"购卡前测试
安装
bash
uvx whichllm@latest # 即装即用(推荐)
uv tool install whichllm # 持久安装
pip install whichllm # pip 安装
使用示例
bash
whichllm # 本机最佳模型
whichllm --gpu "RTX 4090" # 模拟某 GPU
whichllm upgrade "RTX 4090" "RTX 5090" "H100" # 升级对比
whichllm run "qwen 2.5 1.5b gguf" # 直接聊天
whichllm snippet "qwen 7b" # 打印 Python 代码片段
输出示例
#1 Qwen/Qwen3.6-27B 27.8B Q5_K_M score 92.8 27 t/s
#2 Qwen/Qwen3-32B 32.0B Q4_K_M score 83.0 31 t/s
#3 Qwen/Qwen3-30B-A3B 30.0B Q5_K_M score 82.7 102 t/s
数据
- 3.9k Stars · 215 Forks · 210 Commits · v0.5.8
- MIT License
- Python 100%