1
Fastllm: A LLM inference library that runs DeepSeek-V4 with 10GB VRAM
单卡10GB显存就能跑满血DeepSeek-R1-671B，支持全系N卡/A卡甚至国产卡，MOE混合推理更是绝活。
Article URL: https://github.com/ztxz16/fastllm Comments URL: https://news.ycombinator.com/item?id=48728290 Points: 3 # Comments: 0