FlagOS 2.1 — PyTorch-Plugin-FL v0.1.0

Release v0.1.0

Initial Release

Other

• Add the initial Implementation based on FlagOS (#1) by @Hchnr
• Support mm OP on Ascend NPU (#5) by @Hchnr
• Add Ascend Backend Support (#6) by @Hchnr
• Implemented inference and training capabilities for Qwen3-0.6B on the Metax platform. (#4) by @Chenyang409
• Dispatcher Infrastructure Cleanup and Naming Unification (#10) by @Hchnr
• Add triton-ascend patch script and update Ascend installation docs (#12) by @Hchnr
• Fix CPU fallback bottlenecks in inference, reaching 96% of CUDA baseline throughput (#14) by @Hchnr
• Feat(metax): MetaX hybrid backend config, abs kernel, and platform-aware ops tests (#13) by @Chenyang409

Contributors

Thanks to all contributors who made this release possible:

• @Hchnr (6 PRs)
• @Chenyang409 (2 PRs)

---

Generated by Release Notes Generator