Quantized LLM Training in Pure Cuda/c++

The GitHub repository llmq provides an implementation of quantized LLM training in pure CUDA/C++, sparking interest in the tech community, although no discussion has emerged yet.

Snapshot generated from the HN discussion