概览
@xiciyang/inference-quantization
Quantize a model checkpoint and prove it: choose a method (default AWQ+INT4), PTQ-calibrate with a leak-free split, pass an independent accuracy gate and a performance gate through real serving endpoints, auto-degrade on failure, and sign off or roll back — fully audited.
README / ZH
插件文档
目录摘要
Quantize a model checkpoint and prove it: choose a method (default AWQ+INT4), PTQ-calibrate with a leak-free split, pass an independent accuracy gate and a performance gate through real serving endpoints, auto-degrade on failure, and sign off or roll back — fully audited.
dsh.pub 核对固定版本的组合包契约、运行时事实与分发语义；完整 README 请查看源仓库。
在 GitHub 阅读完整 README