1
Pollux – a natively vector quantized LLM with 0.76 bits per parameter
0.76比特/参数的原生量化LLM，用李奇晶格实现极致压缩，性能与体积兼得的新思路。
Article URL: https://github.com/alavicka/pollux Comments URL: https://news.ycombinator.com/item?id=48746067 Points: 1 # Comments: 0