ci: centralize PyTorch test image config

[gyohuangxin]

Summary

• Add a reusable CI config workflow that centralizes the ROCm PyTorch image tags
• Update AITER, Triton, OPUS, tuning, FlashAttention integration, and Triton wheel workflows to read the shared image config
• Keep the Triton workflow path-filter fix so workflow-only changes trigger Triton Test

Validation

• Parsed all workflow YAML files with PyYAML
• git diff --check
• Confirmed no rocm/pytorch:latest references remain under .github/
• Confirmed the ROCm PyTorch 2.10 image tags are only defined in .github/workflows/ci-config.yaml

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:gfx1250-ffm-triton	Run the five-shard gfx1250 FFM Triton test suite
ci:triton-300x	Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang	SGLang integration tests: DeepSeek-R1-MXFP4 accuracy, Qwen 3.5 accuracy
ci:atom	ATOM benchmark: DeepSeek-R1-0528, GPT-OSS-120B
ci:atom_full	ATOM accuracy suite for PR and main models from ATOM models_accuracy.json
ci:vllm	vLLM benchmark: GPT-OSS-120B, DeepSeek-R1-0528, Kimi-K2.5
ci:all	All standard extended tests (excludes ci:atom_full)

Only add ci:atom_full for FlyDSL or Triton upgrades.
Add labels via the sidebar or gh pr edit 4750 --add-label <label>

[Copilot]

Pull request overview

This PR centralizes the ROCm PyTorch Docker image tags used in CI by introducing a reusable workflow that exposes image values as outputs, and then wiring existing CI workflows to consume those outputs instead of hardcoding image tags.

Changes:

• Add .github/workflows/ci-config.yaml reusable workflow to provide shared outputs (PyTorch image tags + Triton wheel artifact name).
• Update multiple CI workflows to needs: ci_config and read needs.ci_config.outputs.* for Docker images (and Triton artifact naming where applicable).
• Adjust Triton workflow path filtering so workflow-only edits (including the new config workflow) still trigger the Triton tests.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
.github/workflows/ci-config.yaml	New reusable workflow exporting centralized ROCm PyTorch image tags and shared artifact naming.
.github/workflows/aiter-test.yaml	Consume centralized image/artifact outputs; update PR path filters to include the shared workflows.
.github/workflows/triton-test.yaml	Consume centralized image/artifact outputs; re-include workflow files in path filters after broad exclusions.
.github/workflows/prepare-triton-wheel.yaml	Make docker-image input required (removing default) to enforce explicit caller configuration.
.github/workflows/opus-test.yaml	Replace hardcoded image with centralized ci_config output.
.github/workflows/operators-tuning.yaml	Replace hardcoded image with centralized ci_config output for container runs.
.github/workflows/tuning-tests.yaml	Move from global DOCKER_IMAGE to per-job ci_config output usage.
.github/workflows/flash_attention_integration.yaml	Add ci-config.yaml to path triggers and use centralized base image output in dependent jobs.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.