← back to paper
arxiv: 2606.10531 · 2 revisions
LC-QAT: Data-Efficient 2-Bit QAT for LLMs via Linear-Constrained Vector Quantization