面向大模型的一站式压缩 Python 包,集成量化与 QEP 等技术,可压缩模型体积并接入 vLLM 加速推理部署。
…) - Support UserDict -based shared KV states when QEP processes transformer blocks one at a time. Documentation - Clarified that OneComp is released under the MIT License
…vice mode synchronization between ModelConfig and QEPConfig. Bug Fix - Stop saving perm in GPTQLinear when actorder=True , as it is unnecessary for inference and causes v
…MqjgJguALXEsBOvdzqu+wjLgAls1+AkuIQAAfP4GFaRkIg6DnGQep6sBe29aQCZ9PlapGH0+7NPgXUx81c4XsqUQ6oEByXpZMfNIxAMie2FEWEAAPiUHwEAAMhNv5deyw2etN83P2eABCbMIiVkwgcaDoAH5ogiYWTA6W2EHth