2026.05.01.post3
Pre-release
Pre-release
·
24 commits
to main
since this release
What's Changed
- feat: pto-isa gdn mega kernel by @fiskrt in #462
- fix insert slice by @zhaozx-cn in #511
- Add interleave mode for split qkv norm rope by @McZyWu in #503
- refactor: separate submodule init and build, remove cuda deps from CI by @fiskrt in #516
- DeepEP: expand moe specifications by @zuje123 in #515
- feat(torch_memory_saver): support 'preload' hook_mode by @Erpim in #463
- [NPU][Feat] Add alltoall mode on deepep by @litmei in #508
New Contributors
- @fiskrt made their first contribution in #462
- @Erpim made their first contribution in #463
- @litmei made their first contribution in #508
Full Changelog: 2026.05.01.post2...2026.05.01.post3
Contributors
Erpim, fiskrt, and 4 other contributors