KernelGenBench v0.2.0

KernelGenBench v0.2.0

Part of FlagOS 2.2.

Changes since v0.1.0 (FlagOS 2.1).

45 pull requests and 1 additional commits.

Source: v0.2.0, commit e3cad63b93af.

New Features

• feat: auto-detect chip type in test_ops.sh (#19) by @factnn
• Add Read the Docs configuration files (#31) by @armstrongttwalker-alt
• feat: multi-chip cuBLAS/CANN backend for Ascend/Hygon/Iluvatar/MUSA/MetaX (#34) by @factnn
• feat: S-Level anti-cheat sandbox with 7-layer defense system (#39) by @factnn
• feat: update analyze.py to v6 with token analysis support (#50) by @factnn
• feat: add MmShapeBench — 50 shapes x 2 dtypes = 100 matmul tests (#51) by @factnn
• feat: add --base-url and --api-key to generate_kernel_and_verify.py (#52) by @factnn
• feat(llm-track): support generic compatible API endpoints (#65) by @factnn

Improvements

• refactor: consolidate sandbox into competition/ package, remove redundancies (#56) by @factnn

Documentation

• docs: separate NVIDIA and domestic chip setup instructions (#26) by @factnn
• docs: sync Chinese README with domestic chip setup instructions (#27) by @factnn
• docs: restructure complete documentation (#32) by @armstrongttwalker-alt
• Docs/restructure complete (#33) by @armstrongttwalker-alt
• Docs/restructure complete (#38) by @armstrongttwalker-alt
• docs: fix hard errors from audit (#40) by @factnn
• docs: fix homepage structure, update docs/ structure (#41) by @armstrongttwalker-alt
• docs: fix 14 content accuracy issues from audit (#42) by @factnn
• Docs/fix homepage (#43) by @armstrongttwalker-alt
• docs: fix glossary, heading, and code example (#44) by @factnn
• Docs/fix homepage (#45) by @armstrongttwalker-alt
• docs: fix doc building warnings (#46) by @armstrongttwalker-alt
• docs: fix zh-CN translation warnings (#47) by @armstrongttwalker-alt
• docs: update zh-CN translation (#48) by @armstrongttwalker-alt
• docs: add anonymous paper PDF and update citation (#54) by @factnn
• docs: fix BibTeX format — double-blind safe (#55) by @factnn
• docs: add competition anti-cheat README (#58) by @factnn
• docs: publish KernelGenBench arXiv link (#62) by @factnn

Testing

• test: add competition anti-cheat test suite with cheat kernels (#57) by @factnn

Bug Fixes

• fix: register empty namespace as triton for non-PyTorch operators (#12) by @factnn
• fix: lower iluvatar dependency versions for CoreX compatibility (#15) by @factnn
• fix: lower ascend dependency versions for NPU container compatibility (#16) by @factnn
• fix: wrap vllm imports with try/except for non-NVIDIA platforms (#18) by @factnn
• fix: repair doubled try/except in vllm import wrappers (#20) by @factnn
• fix: lower metax dependency versions for MUXI container compatibility (#21) by @factnn
• fix: lower hygon dependency versions for DCU container compatibility (#24) by @factnn
• fix: lower musa dependency versions for MUSA container compatibility (#25) by @factnn
• fix: auto-set TORCH_DEVICE_BACKEND_AUTOLOAD=0 on domestic chips (#28) by @factnn
• fix: switch anti-hack Layer 1 from blacklist to torch API whitelist (#36) by @factnn
• fix: add import alias tracking and getattr detection to anti-hack (#37) by @factnn
• fix: support ANTHROPIC_AUTH_TOKEN and base_url for Anthropic providers (#53) by @factnn
• Fix linear and matmul accuracy tests (#66) by @factnn

Other

• chore: add config.yaml with python3 default for agent_bench (#17) by @factnn
• chore: add Apache 2.0 copyright header to all source files (#59) by @tengqm
• Update paper PDF to arxiv version (#60) by @factnn
• Set agent benchmark default max retries to 1 (#63) by @factnn
• Change project license from MIT to Apache 2.0 (e18a477445e0; commit)

Contributors

Thanks to the contributors to the pull requests included in this release.

• @factnn (34 PRs)
• @armstrongttwalker-alt (10 PRs)
• @tengqm

FlagOS 2.1 — KernelGenBench v0.1.0

Release v0.1.0

Initial Release

Other

• Add images for banner and social media links (#1) by @tengqm
• Add images again (#2) by @tengqm
• Docs: update related projects and third-party notices (#3) by @factnn
• Fix: correct Chinese README link (#4) by @factnn
• Fix: table formatting - prevent line wrapping (#5) by @factnn
• Fix: make vllm optional for non-NVIDIA platforms (#8) by @factnn
• Feat: per-platform requirements files (#9) by @factnn
• Docs: update Results with full experiment tables (#10) by @factnn
• Docs: simplify Supported Devices section (#11) by @factnn
• Fix: register empty namespace as triton for non-PyTorch operators (#12) by @factnn
• Fix: lower iluvatar dependency versions for CoreX compatibility (#15) by @factnn
• Fix: lower ascend dependency versions for NPU container compatibility (#16) by @factnn
• Chore: add config.yaml with python3 default for agent_bench (#17) by @factnn
• Fix: wrap vllm imports with try/except for non-NVIDIA platforms (#18) by @factnn
• Feat: auto-detect chip type in test_ops.sh (#19) by @factnn
• Fix: repair doubled try/except in vllm import wrappers (#20) by @factnn
• Fix: lower metax dependency versions for MUXI container compatibility (#21) by @factnn
• Fix: lower hygon dependency versions for DCU container compatibility (#24) by @factnn
• Fix: lower musa dependency versions for MUSA container compatibility (#25) by @factnn
• Docs: separate NVIDIA and domestic chip setup instructions (#26) by @factnn
• Docs: sync Chinese README with domestic chip setup instructions (#27) by @factnn
• Fix: auto-set TORCH_DEVICE_BACKEND_AUTOLOAD=0 on domestic chips (#28) by @factnn
• Add Read the Docs configuration files (#31) by @armstrongttwalker-alt
• Docs: restructure complete documentation (#32) by @armstrongttwalker-alt
• Feat: multi-chip cuBLAS/CANN backend for Ascend/Hygon/Iluvatar/MUSA/MetaX (#34) by @factnn
• Docs/restructure complete (#33) by @armstrongttwalker-alt
• Fix: switch anti-hack Layer 1 from blacklist to torch API whitelist (#36) by @factnn
• Fix: add import alias tracking and getattr detection to anti-hack (#37) by @factnn
• Docs/restructure complete (#38) by @armstrongttwalker-alt
• Feat: S-Level anti-cheat sandbox with 7-layer defense system (#39) by @factnn
• Docs: fix hard errors from audit (#40) by @factnn
• Docs: fix 14 content accuracy issues from audit (#42) by @factnn
• Docs: fix homepage structure, update docs/ structure (#41) by @armstrongttwalker-alt
• Docs/fix homepage (#43) by @armstrongttwalker-alt
• Docs: fix glossary, heading, and code example (#44) by @factnn
• Docs/fix homepage (#45) by @armstrongttwalker-alt
• Docs: fix doc building warnings (#46) by @armstrongttwalker-alt
• Docs: fix zh-CN translation warnings (#47) by @armstrongttwalker-alt
• Docs: update zh-CN translation (#48) by @armstrongttwalker-alt

Contributors

Thanks to all contributors who made this release possible:

• @factnn (27 PRs)
• @armstrongttwalker-alt (10 PRs)
• @tengqm (2 PRs)

---

Generated by Release Notes Generator