fix(ascend): use CommInitAll for single-node TP - #566
Merged
voltjia merged 1 commit intoSep 10, 2026
Conversation
baominghelly
force-pushed
the
fix/ascend-single-node-tp
branch
from
September 10, 2026 03:24
f5f8464 to
561c04c
Compare
baominghelly
marked this pull request as ready for review
September 10, 2026 06:15
voltjia
approved these changes
Sep 10, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
infinicclCommInitAllfor Ascend single-node tensor-parallel communicator initialization.GetUniqueId/CommInitRankpath for platforms that support rank-based initialization.CommInitAllrather thanCommInitRank.Motivation
The current Ascend HCCL backend in InfiniCCL supports
CommInitAll,AllReduce, andCommDestroy, but does not supportGetUniqueIdorCommInitRank. As a result, Ascend TP initialization fails with:This change aligns InfiniLM with the current Ascend InfiniCCL contract. It depends on the Ascend HCCL backend changes tracked in InfiniCCL PR #73.
Type of Change
fix— bug fixTest Results of Involved Models on Supported Platforms
Tested on Ascend 910C with the modern InfiniRT / InfiniOps / InfiniCCL stack and InfiniLM
refactor/adopt-modern-infini-stackat3faa1db.Verified communicator initialization:
Representative inference results after the fix:
FM9G_70B, flash/paged attention, TP8: command completed and generated coherent Chinese output.Qwen3-32B, flash/paged attention, TP4: generated a valid thinking block and final answer.Llama-3.3-70B-Instruct, flash/paged attention, TP8: generated coherent Chinese output.Baichuan2-7B-Chat, flash/paged attention, TP2: generated a coherent factual response.Static test:
Benchmark / Performance Impact
N/A. This is a correctness fix required before Ascend TP inference can initialize.
Notes for Reviewers
initializeCommunicators; this PR only changes the intra-node TP path.CI / ChatOps
CI needs to be triggered manually from the Actions tab, or by a maintainer with
/retestor/test.Checklist
Title, Branch, and Commits
<type>/xxx-yyyy-zzzz.main.fixup!/squash!/wipcommits remain.Scope and Design
General Code Hygiene
C++ Specific
new/delete.csrc/models/llama_legacy/.Python Specific
Testing
Build, CI, and Tooling
_infinilminstall passed on Ascend.Documentation
Security and Safety