Skip to content

Claude Opus 4.6#650

Merged
Oseltamivir merged 2 commits intomainfrom
Oseltamivir-patch-1
Feb 5, 2026
Merged

Claude Opus 4.6#650
Oseltamivir merged 2 commits intomainfrom
Oseltamivir-patch-1

Conversation

@Oseltamivir
Copy link
Copy Markdown
Collaborator

No description provided.

@Oseltamivir Oseltamivir merged commit f5d641c into main Feb 5, 2026
@Oseltamivir Oseltamivir deleted the Oseltamivir-patch-1 branch February 5, 2026 17:58
@Oseltamivir Oseltamivir mentioned this pull request Apr 26, 2026
4 tasks
Oseltamivir added a commit that referenced this pull request Apr 26, 2026
Upstream triton-lang/triton refactored matmul_ogs into matmul.py
(removing routing.py too), but PR #650's fused_moe_triton.py imports
`from triton_kernels.matmul_ogs import matmul_ogs, PrecisionConfig`
and `from triton_kernels.routing import routing`. Those only resolve
against the ROCm fork's RI3.5.x branch.

Pin to ROCm/triton@e491726 (RI3.5.x HEAD, 2025-11-20). Verified at
that SHA:
  * matmul_ogs.py has class PrecisionConfig and def matmul_ogs
  * routing.py exists
  * tensor_details/layout.py exports CDNA4MXScaleLayout
  * target_info.py imports only is_hip / is_hip_cdna3 / is_hip_cdna4
    (no is_hip_gfx1250, which the image's bundled triton rejects)

This matches what AMD's /triton-test/python/triton_kernels/ would have
been on the build host before the release image's cleanup stage.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Development

Successfully merging this pull request may close these issues.

1 participant