Skip to content

mamba-ascendc v0.1.0

Choose a tag to compare

View all tags
@So-cean So-cean released this 10 Aug 09:59
· 5 commits to main since this release
v0.1.0
fb5b945
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Initial public release of the native AscendC Mamba-2 SSD operator.

Highlights:

  • AscendC forward with optional D, z, dt bias/softplus, initial state, and final-state output
  • public-autograd backward with M0 correctness fallback and optimized M1 path
  • PyTorch oracle, Triton-Ascend comparison, A100 baselines, msprof analysis, and Vision-Mamba2 integration
  • reproducible benchmark figures, CPU CI, Apache-2.0 licensing, and third-party provenance
  • one wheel packaging flow for the PyTorch extension and custom OPP

Binary wheels are intentionally not attached because custom OPP binaries depend on the target Ascend SoC and CANN toolchain. Build from source with scripts/build_mamba_ascendc_wheel.sh. See CHANGELOG.md for validated environments and current limitations.

Assets 2
Loading