Update Gluon GEMM-A8W8_BLOCKSCALE README to match nested config layout and supported archs - #4973
Merged
Boss2002n merged 4 commits intoAug 25, 2026
Conversation
…onfig resolution Add resolve_config_dir(), the shared candidate-directory probe from configs/CLAUDE.md section 5 step 1, to gemm_config_utils.py and refactor _get_gemm_config_cached() onto it. Behavior is identical: same candidate order, same winning directory, same default stem, same specialized-file paths. Rewrite the last hand-built config loader, gluon/gemm_a8w8_blockscale.py::_get_config_cached(), on top of resolve_config_dir(..., backend="gluon") so it honors the nested <arch>/gluon/gemm/gemm_a8w8_blockscale/ layout and still falls back to the legacy flat gemm/gluon/ paths until its configs move. Its custom _SUPPORTED_TILES bucket walk is unchanged. Prerequisite for migrating the gemm/gluon config files and for the later MOE resolution unification.
Pure renames (100% similarity), no content changes: the 25 configs/gemm/gluon/<arch>-GEMM-A8W8_BLOCKSCALE*.json and <arch>-GEMM-A8W8_BLOCKSCALE_PRESHUFFLED*.json files move to configs/<arch>/gluon/gemm/<d_type>/ for gfx950 and gfx1250, dropping the arch prefix; each default becomes DEFAULT.json. The resolver (via resolve_config_dir) probes the nested gluon directory first, so these families now resolve from the new location.
Co-authored-by: Boss2002n <58637181+Boss2002n@users.noreply.github.com>
Copilot
AI
changed the title
[WIP] Fix code according to review comment
Update Gluon GEMM-A8W8_BLOCKSCALE README to match nested config layout and supported archs
Aug 24, 2026
Boss2002n
force-pushed
the
satya/unify-gluon-a8w8-blockscale-config-resolution
branch
from
August 24, 2026 23:39
d9ccb7d to
79c2560
Compare
Boss2002n
removed their request for review
August 25, 2026 05:10
Boss2002n
marked this pull request as ready for review
August 25, 2026 05:10
Boss2002n
merged commit Aug 25, 2026
a250b9d
into
satya/unify-gluon-a8w8-blockscale-config-resolution
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Gluon
gemm_a8w8_blockscaledocs still described the legacy flat config path and claimed gfx950-only support, despite config resolution now probing the nested<arch>/gluon/gemm/<family>/layout and the kernel supporting gfx1250.Changes
gfx950 (CDNA4) only→gfx950 (CDNA4), gfx1250, matchingarch_info.is_gluon_avail()and the shippedgfx1250/gluon/...configs.$AITER_TRITON_CONFIGS_PATH/gemm/gluon/gfx950-GEMM-A8W8_BLOCKSCALE[-N=*-K=*].json→ nested$AITER_TRITON_CONFIGS_PATH/<arch>/gluon/gemm/gemm_a8w8_blockscale/{DEFAULT,GEMM-A8W8_BLOCKSCALE-N=*-K=*}.json, reflecting the actual on-disk layout resolved byresolve_config_dir().Scoped to the
GEMM-A8W8_BLOCKSCALEsection only; the unrelatedgemm_a8w8.pydoc block is untouched.