1
Show HN: Turboquant.cpp – Quantize embeddings to 1-4 bits, no training (400 LoC)
仅400行C++代码实现1-4bit嵌入向量量化，无需训练即可极速压缩，性能与精度兼得。
Article URL: https://github.com/RunEdgeAI/turboquant.cpp Comments URL: https://news.ycombinator.com/item?id=48544682 Points: 2 # Comments: 0