Release 0.6.0+triton3.6
Changes since 0.5.0rc2.post1+triton3.6
- Integrates more backends (damoacademy, ARM64 cpu, rpu), and upgraded some backends to Triton 3.6.
- Introduces
tle.pipe, a typed producer/consumer dataflow abstraction for TLE. - Enhances CUDA backend in TLE-Raw.
Documentation
- Add ascend3.5 install readme (#672) by @zhzhcookie
- Update rpu & ascend readme (#681) by @zhzhcookie
- Add rpu readme and update nv/mthreads whl version (#695) by @zhzhcookie
CI/Infrastructure
- Update ci runner yml and readme for docker name (#692) by @zhzhcookie
Dependencies
- Bump docker/setup-buildx-action from 3 to 4 (#671) by @app/dependabot
- Bump actions/upload-artifact from 4 to 7 (#670) by @app/dependabot
Other
- [Ascend] Spec ascend3.5 python code (#637) by @zhzhcookie
- Sync ascend backend, move some path from flir back to third_party/ascend (#660) by @zhzhcookie
- Add AddRmsNormBias triton kernel on ascend (#641) by @small-cat
- Upgrade to flagtree v0.6.0 (#661) by @zhzhcookie
- [CI/CD] Change label and fix default platform (#662) by @zhzhcookie
- Upgrade to flagtree v0.6.0 (#663) by @zhzhcookie
- Build release in CD Runner (#664) by @zhzhcookie
- Guard shared pointer vectorization by alignment (#666) by @sunnycase
- Upgrade to triton 3.5 from TA with commit c8552741 (#635) by @zhzhcookie
- Optimize hint manager lookup (#665) by @sunnycase
- Support enflame change min_bs (#676) by @zhzhcookie
- Fix flagtree spec to flagtree_spec.py (#677) by @zhzhcookie
- [RPU] Add Rhino RPU backend (Huixi Intelligence) (#646) by @xup16
- [MTHREADS] Support TLE Lite primitives (#683) by @Kylin1207
- Fix FP8 WGMMA operand reuse and loop-carried WGMMA waits (#684) by @sunnycase
- Fix device_print on ascend3.5 (#688) by @zhzhcookie
- [DOC] Add TLE compile_hint and interface documentatioin. (#675) by @wangziy1
- [TLE] update for more tle support (#678) by @baoqiliu
- Add 'flagcx' to enabled third-party flags (#689) by @Galaxy1458
- Fix general dot analysis (#691) by @zhzhcookie
- [CI] Stop a wedged launch_kernel dispatch from failing the board smoke (#693) by @xup16
- [DOC] Add RPU backend install guide (English + 中文) (#694) by @xup16
Contributors
Thanks to all contributors who made this release possible:
- @zhzhcookie (15 PRs)
- @sunnycase (3 PRs)
- @xup16 (3 PRs)
- @app/dependabot (2 PRs)
- @Galaxy1458
- @Kylin1207
- @baoqiliu
- @small-cat
- @wangziy1
Generated by Release Notes Generator