11 ms·Fastllm: A LLM inference library that runs DeepSeek-V4 with 10GB VRAM3 points by nogajun 3mo ago