Cherry-pick: Fix regression in Mistral-Large-3-675B (#1304) for v0.19.0#1345
Closed
skavulya wants to merge 8 commits intovllm-project:releases/v0.19.0from
Closed
Cherry-pick: Fix regression in Mistral-Large-3-675B (#1304) for v0.19.0#1345skavulya wants to merge 8 commits intovllm-project:releases/v0.19.0from
skavulya wants to merge 8 commits intovllm-project:releases/v0.19.0from
Conversation
Rename FP8 blockwise compressed tensors scales to match HPU ops Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
Signed-off-by: Kavulya, Soila P <soila.p.kavulya@intel.com>
skavulya
force-pushed
the
skavulya/mistral3-rename-scales-0.19.0
branch
from
c0070da to
f9758c7
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
Cherry-pick for v0.19.0 to fix an FP8 loading/runtime regression for blockwise-compressed tensors on HPU (notably affecting Mistral-Large-3-675B), primarily by aligning scale tensor attribute naming with what HPU FP8 ops expect.
Changes:
- Add scale-attribute aliasing (
*_weight_scale→*_weight_scale_inv) for FP8 block-quantized Linear and MoE paths to match HPU op conventions. - Route FP8 block Linear execution through an HPU block FP8 helper and normalize FP8 linear input/output reshaping.
- Add unit tests covering FP8 block-quantized CompressedTensors Linear and MoE weight post-processing and execution.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
vllm_gaudi/ops/hpu_compressed_tensors.py |
Adds scale aliasing for block quant, updates dequant/apply paths, and adds MoE quant-config helper. |
vllm_gaudi/attention/oot_mla.py |
Ensures dequantized MLA kv projection weights are contiguous to avoid runtime overhead/issues. |
tests/unit_tests/ops/test_hpu_compressed_tensors.py |
Adds new unit tests for FP8 block CompressedTensors Linear and MoE flows. |
✅ CI PassedAll checks passed successfully against the following vllm commit: |
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Signed-off-by: Soila Kavulya <soila.p.kavulya@intel.com>
✅ CI PassedAll checks passed successfully against the following vllm commit: |
Collaborator
|
We want to release tomorrow. This model is not of the highest priority, and this PR didn't go through full QA validation yet. |
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.
Cherry-pick of #1304 for v0.19.0
Renames FP8 blockwise compressed tensors scales to match HPU ops, Fixes regression in https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512 due to #1220 and #1053