DeepSpeed 发布 v0.19.9 补丁版本
原标题:v0.19.9 Patch Release
AI 摘要
DeepSpeed 发布 v0.19.9 补丁版本,包含约 30 项改动,主要涉及 AutoTP、AutoEP、ZeRO 优化器、Triton 内核和 CPU Adam 等组件的修复与增强。新增了面向 vocab-parallel logprobs、entropy 和 KL 的 Triton 内核,为 CPU Adam 加入 RVV 向量化 FP32 内核,并修复了 Triton masked softmax 的数值稳定性问题。此外还修复了文档死链、CI 超时和通信日志性能回退等问题,并有 6 位新贡献者首次参与。
正文节选
## What's Changed * Update version post 0.19.8 release by @loadams in https://github.com/deepspeedai/DeepSpeed/pull/8871 * Update to filter out version.txt from PR gates. by @loadams in https://github.com/deepspeedai/DeepSpeed/pull/8872 * Skip the accelerator-only CPU optimizer tests on the cpu accelerator by @delock in https://github.com/deepspeedai/DeepSpeed/pull/8836 * Keep Yuan's o_proj bias whole on every rank and describe it as replicated by @Achyuthan-S in https://github.com/deepspeed