Skip to content

[ROCm]: test: update HLO references after tmem optimizations (PR4)#4194

Open
cj401-amd wants to merge 5 commits into
AI-Hypercomputer:mainfrom
cj401-amd:cj/tmem-fixes-clean-4-hlo-references
Open

[ROCm]: test: update HLO references after tmem optimizations (PR4)#4194
cj401-amd wants to merge 5 commits into
AI-Hypercomputer:mainfrom
cj401-amd:cj/tmem-fixes-clean-4-hlo-references

Conversation

@cj401-amd

Copy link
Copy Markdown
Collaborator

Summary

Update golden HLO reference files to reflect the changed XLA output after the tmem
optimizations in PRs 1–3. The HLO structure changes due to:

  • Removed ppermute collectives (replaced with slice/pad)
  • Removed redundant sharding constraints
  • Changed attention mask materialization for synthetic data

Updated files:

  • tests/utils/reference_hlo_deepseek3.txt
  • tests/utils/reference_hlo_llama3_8b.txt
  • tests/utils/reference_hlo_qwen3_1.7b.txt

Test plan

  • python3 -m pytest tests/integration/hlo_diff_test.py -v

@cj401-amd cj401-amd force-pushed the cj/tmem-fixes-clean-4-hlo-references branch from 7e7d36b to ba88323 Compare June 18, 2026 22:43
@codecov

codecov Bot commented Jun 18, 2026

Copy link
Copy Markdown

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant