open-source-ai

Hugging Face author maps tensor sensitivity to improve GGUF quantization

A 96-hour experiment across 1,000-plus model variants points to more precise, per-tensor bit allocation for llama.cpp quantization.