CI: Replace nightly image build with S3 aiter wheel installation by gyohuangxin · Pull Request #303 · ROCm/ATOM

gyohuangxin · 2026-03-11T07:19:23Z

Summary

Remove build_atom_image job to eliminate nightly image pull and rebuild overhead
Use rocm/pytorch:latest as the base image instead of rocm/atom-dev:latest
Install the latest amd-aiter wheel from S3 (s3://framework-whls-nightlies/whl-staging/gfx942-gfx950/) at runtime instead of building from source
Install ATOM and other dependencies (lm-eval, hf_transfer, pybind11) directly inside the container

Test plan

Verify the S3 wheel download step finds and installs the latest amd-aiter wheel correctly
Verify ATOM and all dependencies install successfully in the container
Run the full ATOM test matrix to confirm inference and accuracy tests pass

- Remove build_atom_image job that pulled nightly image and rebuilt - Use rocm/pytorch:latest as base image directly - Install latest amd-aiter wheel from S3 bucket at runtime - Install ATOM and dependencies inside the container - Remove fork-specific Dockerfile build logic and Docker Login step

Copilot

Pull request overview

This PR updates the ATOM CI workflow to stop building/pushing a custom nightly Docker image and instead run tests from a base ROCm PyTorch image while installing amd-aiter, ATOM, and Python dependencies at container runtime.

Changes:

Removes the build_atom_image job and related fork/non-fork image handling.
Switches the test container base image to rocm/pytorch:latest.
Adds steps to download/install the latest amd-aiter wheel from S3 and to install ATOM + dependencies inside the running container.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-11T07:24:28Z

.github/workflows/atom-test.yaml

+            pip install awscli 2>/dev/null || true
+


pip install awscli 2>/dev/null || true can leave the container without the aws CLI (e.g., transient pip failure) and then the next aws s3 ... will fail with a less actionable command not found. Consider failing the step if awscli installation fails and/or add an explicit command -v aws check with a clear error message.

Suggested change

pip install awscli 2>/dev/null || true

if ! command -v aws >/dev/null 2>&1; then

echo '=== Installing awscli ==='

if ! pip install awscli; then

echo 'ERROR: Failed to install awscli; cannot download amd-aiter wheel from S3'

exit 1

fi

fi

if ! command -v aws >/dev/null 2>&1; then

echo 'ERROR: aws CLI is not available after installation attempt; cannot download amd-aiter wheel from S3'

exit 1

fi

Copilot · 2026-03-11T07:24:28Z

.github/workflows/atom-test.yaml

+            echo '=== Finding latest amd-aiter wheel from S3 ==='
+            LATEST_WHL=\$(aws s3 ls ${{ env.AITER_S3_BUCKET }}/ --no-sign-request \
+              | grep 'amd_aiter.*\.whl' \
+              | sort -k1,2 \
+              | tail -1 \
+              | awk '{print \$4}')
+
+            if [ -z \"\$LATEST_WHL\" ]; then
+              echo 'ERROR: No amd-aiter wheel found in S3 bucket'
+              exit 1
+            fi
+
+            echo \"Latest wheel: \$LATEST_WHL\"
+            echo '=== Downloading wheel ==='
+            aws s3 cp ${{ env.AITER_S3_BUCKET }}/\$LATEST_WHL /tmp/\$LATEST_WHL --no-sign-request
+
+            echo '=== Uninstalling existing amd-aiter ==='
+            pip uninstall -y amd-aiter || true
+
+            echo '=== Installing amd-aiter from wheel ==='
+            pip install /tmp/\$LATEST_WHL


Selecting and installing the "latest" wheel from S3 makes CI runs non-reproducible and can introduce flakiness (a newer wheel could land between reruns of the same commit). Consider pinning the wheel (e.g., via a version/commit marker file in S3, an explicit env override, or a deterministic naming convention) and logging the resolved wheel’s checksum/version so failures can be reproduced.

Suggested change

echo '=== Finding latest amd-aiter wheel from S3 ==='

LATEST_WHL=\$(aws s3 ls ${{ env.AITER_S3_BUCKET }}/ --no-sign-request \

| grep 'amd_aiter.*\.whl' \

| sort -k1,2 \

| tail -1 \

| awk '{print \$4}')

if [ -z \"\$LATEST_WHL\" ]; then

echo 'ERROR: No amd-aiter wheel found in S3 bucket'

exit 1

fi

echo \"Latest wheel: \$LATEST_WHL\"

echo '=== Downloading wheel ==='

aws s3 cp ${{ env.AITER_S3_BUCKET }}/\$LATEST_WHL /tmp/\$LATEST_WHL --no-sign-request

echo '=== Uninstalling existing amd-aiter ==='

pip uninstall -y amd-aiter || true

echo '=== Installing amd-aiter from wheel ==='

pip install /tmp/\$LATEST_WHL

if [ -n \"\${AITER_WHEEL_NAME:-}\" ]; then

echo '=== Using pinned amd-aiter wheel from AITER_WHEEL_NAME ==='

SELECTED_WHL=\"\$AITER_WHEEL_NAME\"

else

echo '=== Finding latest amd-aiter wheel from S3 ==='

SELECTED_WHL=\$(aws s3 ls ${{ env.AITER_S3_BUCKET }}/ --no-sign-request \

| grep 'amd_aiter.*\.whl' \

| sort -k1,2 \

| tail -1 \

| awk '{print \$4}')

fi

if [ -z \"\$SELECTED_WHL\" ]; then

echo 'ERROR: No amd-aiter wheel found in S3 bucket'

exit 1

fi

echo \"Selected wheel: \$SELECTED_WHL\"

echo '=== Downloading wheel ==='

aws s3 cp ${{ env.AITER_S3_BUCKET }}/\$SELECTED_WHL /tmp/\$SELECTED_WHL --no-sign-request

echo '=== Wheel SHA256 checksum ==='

sha256sum /tmp/\$SELECTED_WHL || echo 'WARNING: sha256sum command failed'

echo '=== Uninstalling existing amd-aiter ==='

pip uninstall -y amd-aiter || true

echo '=== Installing amd-aiter from wheel ==='

pip install /tmp/\$SELECTED_WHL

Copilot · 2026-03-11T07:24:29Z

.github/workflows/atom-test.yaml

+            echo \"Latest wheel: \$LATEST_WHL\"
+            echo '=== Downloading wheel ==='
+            aws s3 cp ${{ env.AITER_S3_BUCKET }}/\$LATEST_WHL /tmp/\$LATEST_WHL --no-sign-request
+
+            echo '=== Uninstalling existing amd-aiter ==='
+            pip uninstall -y amd-aiter || true
+
+            echo '=== Installing amd-aiter from wheel ==='
+            pip install /tmp/\$LATEST_WHL
+


The workflow downloads and installs a wheel from a public S3 location with --no-sign-request but does not perform any integrity verification before pip install. To reduce supply-chain risk, consider fetching a corresponding checksum/signature (e.g., .sha256) and verifying it before installation, or using a signed URL / authenticated access.

…runner - Add aws-actions/configure-aws-credentials step with IAM role - Download aiter wheel on the runner (not inside container) - Copy wheel into container via docker cp for installation

- Remove AWS credentials and S3 bucket configuration - Download latest aiter wheel from ROCm/aiter CI artifacts via GitHub API - Use GITHUB_TOKEN for cross-repo artifact access

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 2 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-11T08:36:24Z

.github/workflows/atom-test.yaml

+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN rm -rf /app/aiter-test
-          RUN git clone https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
-
-          RUN echo "=== ATOM version AFTER installation ===" && pip show atom || true
-          EOF
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="

-      - name: Build Docker image for forked repo
-        if: (matrix.run_on_pr == true || github.event_name != 'pull_request') && github.event.pull_request.head.repo.fork
-        run: |
-          docker build --pull --network=host \
-            --no-cache \
-            -t atom_test:ci \
-            -f Dockerfile.mod .
+          ARTIFACT_JSON=$(gh api "repos/ROCm/aiter/actions/artifacts?per_page=100" \
+            --jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | sort_by(.created_at) | last')


The workflow uses ${{ secrets.GITHUB_TOKEN }} (scoped to this repo) to call repos/ROCm/aiter/actions/artifacts/.... GITHUB_TOKEN cannot access private resources (including Actions artifacts) in a different repository, so this step will fail on push/schedule and on PRs (especially forks). Consider switching to the S3 wheel source described in the PR, or minting a token with actions:read on ROCm/aiter (GitHub App/PAT) and adding a fork-safe fallback/skip path.

Copilot · 2026-03-11T08:36:24Z

.github/workflows/atom-test.yaml

+      - name: Download latest aiter wheel from CI artifacts
+        if: matrix.run_on_pr == true || github.event_name != 'pull_request'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN rm -rf /app/aiter-test
-          RUN git clone https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
-
-          RUN echo "=== ATOM version AFTER installation ===" && pip show atom || true
-          EOF
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="


PR description says the latest amd-aiter wheel is installed from S3 (s3://framework-whls-nightlies/...), but the workflow now downloads a GitHub Actions artifact from ROCm/aiter. If S3 is the intended source of truth, this step should be updated to match (or the PR description updated) to avoid confusion about provenance and required credentials.

Copilot

Pull request overview

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

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-11T14:43:49Z

.github/workflows/atom-test.yaml

+            | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | sort_by(.created_at) | last')
+
+          ARTIFACT_NAME=$(echo "$ARTIFACT_JSON" | jq -r '.name')
+          ARTIFACT_ID=$(echo "$ARTIFACT_JSON" | jq -r '.id')


The workflow selects the most recently created aiter-whl-main artifact and installs it without any version/commit pinning or integrity verification. This makes CI non-reproducible and increases supply-chain risk if an unexpected artifact is published. Consider pinning to a specific aiter commit/SHA (or a versioned wheel path), and/or verifying a checksum/signature before installing.

…isting

Copilot

Pull request overview

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

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-12T03:27:14Z

.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          # Search Aiter Test workflow runs on main branch for one that has an aiter-whl artifact
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")


This uses the workflow run’s GITHUB_TOKEN to download artifacts from a different repository (ROCm/aiter). The token is scoped to the current repo and typically cannot access cross-repo Actions artifacts, which will commonly fail with 403/404. Use a dedicated secret (PAT or GitHub App token) that has actions:read on ROCm/aiter, or switch to the PR-described S3 wheel source to avoid cross-repo artifact auth entirely.

Copilot · 2026-03-12T03:27:15Z

.github/workflows/atom-test.yaml

+            ARTIFACT_JSON=$(curl -s -H "$AUTH_HEADER" \
+              "$API_URL/repos/ROCm/aiter/actions/runs/$RUN_ID/artifacts" \
+              | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | first')


This uses the workflow run’s GITHUB_TOKEN to download artifacts from a different repository (ROCm/aiter). The token is scoped to the current repo and typically cannot access cross-repo Actions artifacts, which will commonly fail with 403/404. Use a dedicated secret (PAT or GitHub App token) that has actions:read on ROCm/aiter, or switch to the PR-described S3 wheel source to avoid cross-repo artifact auth entirely.

Copilot · 2026-03-12T03:27:15Z

.github/workflows/atom-test.yaml

+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do


The script selects the first run (most recent by API ordering) that contains a non-expired artifact, but does not require the run to be status=completed and conclusion=success. This can pick artifacts from failed or in-progress runs and lead to installing a broken wheel. Filter workflow runs to completed+successful before searching artifacts (or validate run conclusion before accepting the artifact).

Suggested change

for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do

for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[] | select(.status=="completed" and .conclusion=="success") | .id'); do

Copilot · 2026-03-12T03:27:15Z

.github/workflows/atom-test.yaml

+              echo "Found artifact in run $RUN_ID: $ARTIFACT_NAME (ID: $ARTIFACT_ID)"
+              break
+            fi
+          done


The script selects the first run (most recent by API ordering) that contains a non-expired artifact, but does not require the run to be status=completed and conclusion=success. This can pick artifacts from failed or in-progress runs and lead to installing a broken wheel. Filter workflow runs to completed+successful before searching artifacts (or validate run conclusion before accepting the artifact).

Copilot · 2026-03-12T03:27:15Z

.github/workflows/atom-test.yaml

+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do


This step assumes jq and unzip are present on the runner host. Previously, jq was installed inside the Docker image, but now parsing/downloading happens before the container is started. Add an explicit dependency-install step for the runner (or rewrite parsing to avoid jq, e.g., using python -c), otherwise this will fail on self-hosted runners that don’t preinstall these tools.

Copilot · 2026-03-12T03:27:15Z

.github/workflows/atom-test.yaml

+          curl -s -L -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/artifacts/$ARTIFACT_ID/zip" \
+            -o /tmp/aiter-whl.zip
+          unzip -o /tmp/aiter-whl.zip -d /tmp/aiter-whl


This step assumes jq and unzip are present on the runner host. Previously, jq was installed inside the Docker image, but now parsing/downloading happens before the container is started. Add an explicit dependency-install step for the runner (or rewrite parsing to avoid jq, e.g., using python -c), otherwise this will fail on self-hosted runners that don’t preinstall these tools.

Copilot · 2026-03-12T03:27:16Z

.github/workflows/atom-test.yaml

+          AITER_WHL=$(ls /tmp/aiter-whl/amd_aiter*.whl 2>/dev/null | head -1)
+          if [ -z "$AITER_WHL" ]; then
+            echo "ERROR: No amd_aiter wheel found in artifact"
+            ls -la /tmp/aiter-whl/
+            exit 1
+          fi
+


If the artifact ever contains multiple matching wheels, head -1 is not guaranteed to pick the newest/desired build. Prefer selecting deterministically (e.g., version-sort then take the highest) or enforce that exactly one wheel is present and fail otherwise.

Suggested change

AITER_WHL=$(ls /tmp/aiter-whl/amd_aiter*.whl 2>/dev/null | head -1)

if [ -z "$AITER_WHL" ]; then

echo "ERROR: No amd_aiter wheel found in artifact"

ls -la /tmp/aiter-whl/

exit 1

fi

AITER_WHL_CANDIDATES=$(ls -1 /tmp/aiter-whl/amd_aiter*.whl 2>/dev/null | sort -V || true)

if [ -z "$AITER_WHL_CANDIDATES" ]; then

echo "ERROR: No amd_aiter wheel found in artifact"

ls -la /tmp/aiter-whl/

exit 1

fi

AITER_WHL=$(echo "$AITER_WHL_CANDIDATES" | tail -n 1)

Copilot · 2026-03-12T03:27:16Z

.github/workflows/atom-test.yaml

+          fi
+
+          echo "Downloaded wheel: $AITER_WHL"
+          echo "AITER_WHL_PATH=$AITER_WHL" >> $GITHUB_ENV


$GITHUB_ENV should be quoted when redirecting to avoid issues with unexpected whitespace/shell expansion. Use a quoted redirect target (and consider using printf for robustness).

Suggested change

echo "AITER_WHL_PATH=$AITER_WHL" >> $GITHUB_ENV

echo "AITER_WHL_PATH=$AITER_WHL" >> "$GITHUB_ENV"

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 2 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-03-12T06:01:24Z

.github/workflows/atom-test.yaml

+          WHL_NAME=$(basename "${{ env.AITER_WHL_PATH }}")
+          docker cp "${{ env.AITER_WHL_PATH }}" "$CONTAINER_NAME:/tmp/$WHL_NAME"


AITER_WHL_PATH is set via $GITHUB_ENV, but later referenced using the ${{ env.AITER_WHL_PATH }} expression context. Values written to $GITHUB_ENV are available as shell environment variables in subsequent steps (e.g., $AITER_WHL_PATH), but are not reliably available through the env expression context, which can make basename/docker cp run with an empty path. Use the runtime shell env var (or step outputs) instead of ${{ env.* }} here.

Suggested change

WHL_NAME=$(basename "${{ env.AITER_WHL_PATH }}")

docker cp "${{ env.AITER_WHL_PATH }}" "$CONTAINER_NAME:/tmp/$WHL_NAME"

WHL_NAME=$(basename "$AITER_WHL_PATH")

docker cp "$AITER_WHL_PATH" "$CONTAINER_NAME:/tmp/$WHL_NAME"

Copilot · 2026-03-12T06:01:24Z

.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          # Search Aiter Test workflow runs on main branch for one that has an aiter-whl artifact
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do
+            ARTIFACT_JSON=$(curl -s -H "$AUTH_HEADER" \
+              "$API_URL/repos/ROCm/aiter/actions/runs/$RUN_ID/artifacts" \
+              | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | first')
+
+            if [ "$ARTIFACT_JSON" != "null" ] && [ -n "$ARTIFACT_JSON" ]; then
+              ARTIFACT_ID=$(echo "$ARTIFACT_JSON" | jq -r '.id')
+              ARTIFACT_NAME=$(echo "$ARTIFACT_JSON" | jq -r '.name')
+              echo "Found artifact in run $RUN_ID: $ARTIFACT_NAME (ID: $ARTIFACT_ID)"
+              break
+            fi
+          done

-      - name: Build Docker image for forked repo
-        if: (matrix.run_on_pr == true || github.event_name != 'pull_request') && github.event.pull_request.head.repo.fork
-        run: |
-          docker build --pull --network=host \
-            --no-cache \
-            -t atom_test:ci \
-            -f Dockerfile.mod .
+          if [ -z "$ARTIFACT_ID" ] || [ "$ARTIFACT_ID" = "null" ]; then
+            echo "ERROR: No aiter-whl-main artifact found in recent Aiter Test runs"
+            exit 1
+          fi
+
+          echo "=== Downloading artifact ==="
+          mkdir -p /tmp/aiter-whl
+          curl -s -L -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/artifacts/$ARTIFACT_ID/zip" \
+            -o /tmp/aiter-whl.zip
+          unzip -o /tmp/aiter-whl.zip -d /tmp/aiter-whl


This step attempts to download a workflow artifact from the ROCm/aiter repository using ${{ secrets.GITHUB_TOKEN }}. The workflow token is scoped to the current repository and typically cannot access private/collaborator-only resources (including Actions artifacts) in other repositories; this is likely to fail with 403s, especially on PRs from forks where secrets/PATs aren’t available. Consider switching to the PR-described S3 wheel source (or another publicly readable location), or use a dedicated PAT with explicit access and a fork-safe fallback behavior.

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 6 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do


.github/workflows/atom-test.yaml

+      - name: Download latest aiter wheel from CI artifacts
+        if: matrix.run_on_pr == true || github.event_name != 'pull_request'
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN rm -rf /app/aiter-test
-          RUN git clone https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-          
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+


.github/workflows/atom-test.yaml


 env:
-  ATOM_BASE_NIGTHLY_IMAGE: rocm/atom-dev:latest
+  ATOM_BASE_IMAGE: rocm/pytorch:latest


.github/workflows/atom-test.yaml

  atom:
-    needs: [pre-checks, build_atom_image]
+    needs: [pre-checks]
    name: ATOM Test


.github/workflows/atom-test.yaml

+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+
+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"


.github/workflows/atom-test.yaml

+          AITER_TEST_WORKFLOW_ID=179476100
+


The build_atom_image job is no longer needed since we now install aiter from CI artifact wheels and ATOM/dependencies at runtime inside the container.

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 5 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+


.github/workflows/atom-test.yaml

+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do


.github/workflows/atom-test.yaml

+          AITER_TEST_WORKFLOW_ID=179476100
+


.github/workflows/atom-test.yaml

+      - name: Find and download latest aiter wheel
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN wget https://github.com/stedolan/jq/releases/download/jq-1.7/jq-linux64 -O jq
-          RUN chmod +x jq
-          RUN mv jq /usr/local/bin/jq
-          RUN rm -rf /app/aiter-test
-          RUN git clone --depth 1 https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
-
-          RUN echo "=== ATOM version AFTER installation ===" && pip show atom || true
-          EOF
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="


.github/workflows/atom-test.yaml

  GITHUB_REPO_URL: ${{ github.event.pull_request.head.repo.clone_url || 'https://github.com/ROCm/ATOM.git' }}
  GITHUB_COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.event.head_commit.id }}


Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 7 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

-
-
-          docker run -dt --device=/dev/kfd $DEVICE_FLAG \
+          docker run -dt --pull always --device=/dev/kfd $DEVICE_FLAG \


.github/workflows/atom-test.yaml

+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")


.github/workflows/atom-test.yaml

+      - name: Find and download latest aiter wheel
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN wget https://github.com/stedolan/jq/releases/download/jq-1.7/jq-linux64 -O jq
-          RUN chmod +x jq
-          RUN mv jq /usr/local/bin/jq
-          RUN rm -rf /app/aiter-test
-          RUN git clone --depth 1 https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
-
-          RUN echo "=== ATOM version AFTER installation ===" && pip show atom || true
-          EOF
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+


.github/workflows/atom-test.yaml

+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+


.github/workflows/atom-test.yaml

+          AITER_TEST_WORKFLOW_ID=179476100
+


.github/workflows/atom-test.yaml

  atom:
-    needs: [pre-checks, build_atom_image]
+    needs: [pre-checks, download_aiter_wheel]
    name: ATOM Test
    strategy:


.github/workflows/atom-test.yaml


 env:
-  ATOM_BASE_NIGTHLY_IMAGE: rocm/atom-dev:latest
+  ATOM_BASE_IMAGE: rocm/pytorch:latest


Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 6 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

-
-
-          docker run -dt --device=/dev/kfd $DEVICE_FLAG \
+          docker run -dt --pull always --device=/dev/kfd $DEVICE_FLAG \


.github/workflows/atom-test.yaml

  GITHUB_REPO_URL: ${{ github.event.pull_request.head.repo.clone_url || 'https://github.com/ROCm/ATOM.git' }}
  GITHUB_COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.event.head_commit.id }}


.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+


.github/workflows/atom-test.yaml

+      - name: Find and download latest aiter wheel
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN wget https://github.com/stedolan/jq/releases/download/jq-1.7/jq-linux64 -O jq
-          RUN chmod +x jq
-          RUN mv jq /usr/local/bin/jq
-          RUN rm -rf /app/aiter-test
-          RUN git clone --depth 1 https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
-
-          RUN echo "=== ATOM version AFTER installation ===" && pip show atom || true
-          EOF
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+
+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100


.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+


.github/workflows/atom-test.yaml


 env:
-  ATOM_BASE_NIGTHLY_IMAGE: rocm/atom-dev:latest
+  ATOM_BASE_IMAGE: rocm/pytorch:latest


Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 5 comments.

Comments suppressed due to low confidence (1)

.github/workflows/atom-test.yaml:273

docker run mounts /workspace twice (both ${GITHUB_WORKSPACE:-$PWD} and ${{ github.workspace }}) and also sets -w /workspace twice. This duplication is easy to miss when editing the command and can lead to confusion about which path is authoritative; consider removing the duplicate -v/-w entries.

          docker run -dt --pull always --device=/dev/kfd $DEVICE_FLAG \
          -v "${GITHUB_WORKSPACE:-$PWD}":/workspace \
          $MODEL_MOUNT \
          -w /workspace \
          --ipc=host --group-add video \
          --shm-size=16G \
          --privileged \
          --cap-add=SYS_PTRACE \
          -e HF_TOKEN="${HF_TOKEN:-}" \
          --env-file /tmp/env_file.txt \
          --security-opt seccomp=unconfined \
          --ulimit memlock=-1 \
          --ulimit stack=67108864 \
          -e ATOM_DISABLE_MMAP=true \
          -v "${{ github.workspace }}:/workspace" \
          -w /workspace \
          --name "$CONTAINER_NAME" \

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

+      - name: Find and download latest aiter wheel
        run: |
-          cat <<EOF > Dockerfile.mod
-          FROM ${{ env.ATOM_BASE_NIGTHLY_IMAGE }}
-          RUN pip install -U lm-eval[api]
-          RUN pip show lm-eval || true
-          RUN pip install hf_transfer
-          RUN pip show hf_transfer || true
-          RUN echo "=== Aiter version BEFORE uninstall ===" && pip show amd-aiter || true
-          RUN pip uninstall -y amd-aiter
-          RUN pip install --upgrade "pybind11>=3.0.1"
-          RUN pip show pybind11
-          RUN wget https://github.com/stedolan/jq/releases/download/jq-1.7/jq-linux64 -O jq
-          RUN chmod +x jq
-          RUN mv jq /usr/local/bin/jq
-          RUN rm -rf /app/aiter-test
-          RUN git clone --depth 1 https://github.com/ROCm/aiter.git /app/aiter-test && \\
-              cd /app/aiter-test && \\
-              git checkout HEAD && \\
-              git submodule sync && git submodule update --init --recursive && \\
-              MAX_JOBS=64 PREBUILD_KERNELS=0 GPU_ARCHS=gfx950 python3 setup.py develop
-          RUN echo "=== Aiter version AFTER installation ===" && pip show amd-aiter || true
-
-          RUN echo "=== ATOM version BEFORE uninstall ===" && pip show atom || true
-          RUN pip uninstall -y atom
-          RUN rm -rf /app/ATOM
-          RUN git clone ${{ env.GITHUB_REPO_URL }} /app/ATOM && \\
-              cd /app/ATOM && \\
-              git checkout ${{ env.GITHUB_COMMIT_SHA }} && \\
-              pip install -e .
-
-          RUN echo "=== ATOM version AFTER installation ===" && pip show atom || true
-          EOF
+          set -euo pipefail
+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+
+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+


.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do
+            ARTIFACT_JSON=$(curl -s -H "$AUTH_HEADER" \
+              "$API_URL/repos/ROCm/aiter/actions/runs/$RUN_ID/artifacts" \
+              | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | first')
+
+            if [ "$ARTIFACT_JSON" != "null" ] && [ -n "$ARTIFACT_JSON" ]; then
+              ARTIFACT_ID=$(echo "$ARTIFACT_JSON" | jq -r '.id')
+              ARTIFACT_NAME=$(echo "$ARTIFACT_JSON" | jq -r '.name')
+              echo "Found artifact in run $RUN_ID: $ARTIFACT_NAME (ID: $ARTIFACT_ID)"
+              break
+            fi
+          done

-      - name: Build Docker image 
-        if: ${{ !github.event.pull_request.head.repo.fork }}
-        run: |
-          docker build --pull --network=host \
-            --no-cache \
-            -t atom_test:ci \
-            -f Dockerfile.mod .
+          if [ -z "$ARTIFACT_ID" ] || [ "$ARTIFACT_ID" = "null" ]; then
+            echo "ERROR: No aiter-whl-main artifact found in recent Aiter Test runs"
+            exit 1
+          fi

-      - name: Push Docker image
-        if: ${{ !github.event.pull_request.head.repo.fork }}
-        run: |
-          IMAGE_TAG=rocm/atom-dev:pre-build-${{ env.GITHUB_COMMIT_SHA }}
-          docker tag atom_test:ci $IMAGE_TAG
-          echo "${{ secrets.DOCKER_PASSWORD }}" | docker login -u ${{ secrets.DOCKER_USERNAME }} --password-stdin
-          docker push $IMAGE_TAG
+          echo "=== Downloading artifact ==="
+          mkdir -p aiter-whl
+          curl -s -L -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/artifacts/$ARTIFACT_ID/zip" \
+            -o aiter-whl.zip


.github/workflows/atom-test.yaml

+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")


.github/workflows/atom-test.yaml

+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do
+            ARTIFACT_JSON=$(curl -s -H "$AUTH_HEADER" \
+              "$API_URL/repos/ROCm/aiter/actions/runs/$RUN_ID/artifacts" \
+              | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | first')
+
+            if [ "$ARTIFACT_JSON" != "null" ] && [ -n "$ARTIFACT_JSON" ]; then
+              ARTIFACT_ID=$(echo "$ARTIFACT_JSON" | jq -r '.id')
+              ARTIFACT_NAME=$(echo "$ARTIFACT_JSON" | jq -r '.name')
+              echo "Found artifact in run $RUN_ID: $ARTIFACT_NAME (ID: $ARTIFACT_ID)"
+              break
+            fi
+          done


.github/workflows/atom-test.yaml

  GITHUB_REPO_URL: ${{ github.event.pull_request.head.repo.clone_url || 'https://github.com/ROCm/ATOM.git' }}
  GITHUB_COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.event.head_commit.id }}


Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+


.github/workflows/atom-test.yaml

+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+
+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do
+            ARTIFACT_JSON=$(curl -s -H "$AUTH_HEADER" \
+              "$API_URL/repos/ROCm/aiter/actions/runs/$RUN_ID/artifacts" \
+              | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | first')
+


.github/workflows/atom-test.yaml


 env:
-  ATOM_BASE_NIGHTLY_IMAGE: rocm/atom-dev:latest
+  ATOM_BASE_IMAGE: rocm/pytorch:latest


.github/workflows/atom-test.yaml

-
-
-          docker run -dt --device=/dev/kfd $DEVICE_FLAG \
+          docker run -dt --pull always --device=/dev/kfd $DEVICE_FLAG \


Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

+  download_aiter_wheel:
+    if: ${{ needs.check-signal.result == 'success' && (!github.event.pull_request || github.event.pull_request.draft == false) }}
    needs: [check-signal]
-    name: Build ATOM image
-    runs-on: build-only-atom
+    name: Download aiter wheel


.github/workflows/atom-test.yaml

+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")
+
+          ARTIFACT_ID=""
+          ARTIFACT_NAME=""
+          for RUN_ID in $(echo "$RUNS" | jq -r '.workflow_runs[].id'); do
+            ARTIFACT_JSON=$(curl -s -H "$AUTH_HEADER" \
+              "$API_URL/repos/ROCm/aiter/actions/runs/$RUN_ID/artifacts" \
+              | jq '[.artifacts[] | select(.name | startswith("aiter-whl-main")) | select(.expired == false)] | first')


.github/workflows/atom-test.yaml

+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+
+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")


.github/workflows/atom-test.yaml


 env:
-  ATOM_BASE_NIGHTLY_IMAGE: rocm/atom-dev:latest
+  ATOM_BASE_IMAGE: rocm/pytorch:latest


Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 9 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

.github/workflows/atom-test.yaml

+          echo "=== Finding latest aiter-whl-main artifact from ROCm/aiter ==="
+
+          API_URL="https://api.github.com"
+          AUTH_HEADER="Authorization: token ${{ secrets.GITHUB_TOKEN }}"
+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")


.github/workflows/atom-test.yaml

+          AITER_TEST_WORKFLOW_ID=179476100
+
+          RUNS=$(curl -s -H "$AUTH_HEADER" \
+            "$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")


.github/workflows/atom-test.yaml

+        with:
+          name: aiter-whl
+          path: /tmp/aiter-whl



.github/workflows/atom-test.yaml

-
-
-          docker run -dt --device=/dev/kfd $DEVICE_FLAG \
+          docker run -dt --pull always --device=/dev/kfd $DEVICE_FLAG \


.github/workflows/atom-test.yaml

+            echo '=== Installing amd-aiter from wheel ==='
+            pip install /tmp/$WHL_NAME
+


.github/workflows/atom-test.yaml

-    name: Build ATOM image
-    runs-on: build-only-atom
+    name: Download aiter wheel
+    runs-on: ubuntu-latest


.github/workflows/atom-test.yaml

+
+      - name: Upload aiter wheel
+        uses: actions/upload-artifact@v4
+        with:
+          name: aiter-whl
+          path: aiter-whl/amd_aiter*.whl


.github/workflows/atom-test.yaml

+          AITER_WHL=$(ls -t aiter-whl/amd_aiter*.whl 2>/dev/null | head -1)
+          if [ -z "$AITER_WHL" ]; then
+            echo "ERROR: No amd_aiter wheel found in artifact"
+            ls -la aiter-whl/
+            exit 1


.github/workflows/atom-test.yaml

+          AITER_WHL=$(ls -t /tmp/aiter-whl/amd_aiter*.whl 2>/dev/null | head -1)
+          if [ -z "$AITER_WHL" ]; then
+            echo "ERROR: No amd_aiter wheel found"
+            ls -la /tmp/aiter-whl/
+            exit 1
+          fi
+
+          echo "=== Copying wheel into container ==="
+          WHL_NAME=$(basename "$AITER_WHL")
+          docker cp "$AITER_WHL" "$CONTAINER_NAME:/tmp/$WHL_NAME"
+
+          docker exec "$CONTAINER_NAME" bash -lc "
+            set -euo pipefail
+            echo '=== Uninstalling existing amd-aiter ==='
+            pip uninstall -y amd-aiter || true
+
+            echo '=== Installing amd-aiter from wheel ==='
+            pip install /tmp/$WHL_NAME


Copilot AI review requested due to automatic review settings March 11, 2026 07:19

Copilot started reviewing on behalf of gyohuangxin March 11, 2026 07:20 View session

Copilot AI reviewed Mar 11, 2026

View reviewed changes

gyohuangxin added 2 commits March 11, 2026 15:52

CI: Fix S3 access by adding AWS credentials and downloading wheel on …

760db33

…runner - Add aws-actions/configure-aws-credentials step with IAM role - Download aiter wheel on the runner (not inside container) - Copy wheel into container via docker cp for installation

CI: Switch from S3 to GitHub artifact for aiter wheel download

205ceba

- Remove AWS credentials and S3 bucket configuration - Download latest aiter wheel from ROCm/aiter CI artifacts via GitHub API - Use GITHUB_TOKEN for cross-repo artifact access

Copilot AI review requested due to automatic review settings March 11, 2026 08:31

Copilot started reviewing on behalf of gyohuangxin March 11, 2026 08:32 View session

Copilot AI reviewed Mar 11, 2026

View reviewed changes

gyohuangxin added 2 commits March 11, 2026 17:37

Fix: replace gh CLI with curl for artifact download

959e4a6

Fix: increase pip timeout and retries for flaky network

45319e2

Copilot AI review requested due to automatic review settings March 11, 2026 14:33

Copilot started reviewing on behalf of gyohuangxin March 11, 2026 14:35 View session

Copilot AI reviewed Mar 11, 2026

View reviewed changes

gyohuangxin added 2 commits March 12, 2026 10:38

Fix: search aiter wheel artifact by workflow runs instead of global l…

b2fb8a9

…isting

Fix: search aiter wheel from main branch push runs only

f10c9fe

Copilot AI review requested due to automatic review settings March 12, 2026 03:24

Copilot AI reviewed Mar 12, 2026

View reviewed changes

gyohuangxin added 2 commits March 12, 2026 12:14

Fix: remove permissions block that breaks reusable workflow

79ea792

Merge branch 'main' into ci/use-s3-aiter-wheel

b88cb10

Copilot AI review requested due to automatic review settings March 12, 2026 05:56

Copilot started reviewing on behalf of gyohuangxin March 12, 2026 05:57 View session

Copilot AI reviewed Mar 12, 2026

View reviewed changes

Merge branch 'main' into ci/use-s3-aiter-wheel

f016493

gyohuangxin marked this pull request as draft March 12, 2026 08:56

gyohuangxin marked this pull request as ready for review March 13, 2026 02:50

Copilot AI review requested due to automatic review settings March 13, 2026 02:50

Copilot started reviewing on behalf of gyohuangxin March 13, 2026 02:51 View session

Copilot AI reviewed Mar 13, 2026

View reviewed changes

Remove build_atom_image job

eaae9dd

The build_atom_image job is no longer needed since we now install aiter from CI artifact wheels and ATOM/dependencies at runtime inside the container.

Copilot AI reviewed Mar 14, 2026

View reviewed changes

gyohuangxin added 2 commits March 16, 2026 00:35

Merge branch 'main' into ci/use-s3-aiter-wheel

dac98fb

Merge branch 'main' into ci/use-s3-aiter-wheel

54249f9

Copilot AI review requested due to automatic review settings March 16, 2026 06:18

Copilot started reviewing on behalf of gyohuangxin March 16, 2026 06:19 View session

Copilot AI reviewed Mar 16, 2026

View reviewed changes

gyohuangxin added 2 commits March 16, 2026 21:58

Merge branch 'main' into ci/use-s3-aiter-wheel

0e645f8

Merge branch 'main' into ci/use-s3-aiter-wheel

0ba131f

Copilot AI review requested due to automatic review settings March 16, 2026 17:19

Copilot started reviewing on behalf of gyohuangxin March 16, 2026 17:19 View session

Copilot AI reviewed Mar 16, 2026

View reviewed changes

CI: Print installed package versions after setup

a366f00

Copilot AI review requested due to automatic review settings March 17, 2026 07:56

gyohuangxin force-pushed the ci/use-s3-aiter-wheel branch from a5eaede to a366f00 Compare March 17, 2026 07:56

Copilot started reviewing on behalf of gyohuangxin March 17, 2026 07:57 View session

Copilot AI reviewed Mar 17, 2026

View reviewed changes

gyohuangxin added 2 commits March 18, 2026 10:30

Merge branch 'main' into ci/use-s3-aiter-wheel

5009201

Merge branch 'main' into ci/use-s3-aiter-wheel

bb79c36

Copilot AI review requested due to automatic review settings March 18, 2026 06:07

Copilot started reviewing on behalf of gyohuangxin March 18, 2026 06:08 View session

Copilot AI reviewed Mar 18, 2026

View reviewed changes

gyohuangxin added 2 commits March 18, 2026 23:34

Merge branch 'main' into ci/use-s3-aiter-wheel

e04551c

Merge branch 'main' into ci/use-s3-aiter-wheel

f6b9b8f

Copilot AI review requested due to automatic review settings March 19, 2026 07:20

gyohuangxin added 2 commits March 19, 2026 15:22

Update atom-test.yaml

7374da5

Update atom-test.yaml

9d044d0

Copilot AI reviewed Mar 19, 2026

View reviewed changes

Update atom-test.yaml

4fcc8ea

Copilot AI review requested due to automatic review settings March 19, 2026 08:35

Copilot AI reviewed Mar 19, 2026

View reviewed changes

-            pip install awscli 2>/dev/null || true
+            if ! command -v aws >/dev/null 2>&1; then
+              echo '=== Installing awscli ==='
+              if ! pip install awscli; then
+                echo 'ERROR: Failed to install awscli; cannot download amd-aiter wheel from S3'
+                exit 1
+              fi
+            fi
+            if ! command -v aws >/dev/null 2>&1; then
+              echo 'ERROR: aws CLI is not available after installation attempt; cannot download amd-aiter wheel from S3'
+              exit 1
+            fi

	for RUN_ID in $(echo "$RUNS" \| jq -r '.workflow_runs[].id'); do
	for RUN_ID in $(echo "$RUNS" \| jq -r '.workflow_runs[] \| select(.status=="completed" and .conclusion=="success") \| .id'); do

	echo "AITER_WHL_PATH=$AITER_WHL" >> $GITHUB_ENV
	echo "AITER_WHL_PATH=$AITER_WHL" >> "$GITHUB_ENV"

		WHL_NAME=$(basename "${{ env.AITER_WHL_PATH }}")
		docker cp "${{ env.AITER_WHL_PATH }}" "$CONTAINER_NAME:/tmp/$WHL_NAME"

		GITHUB_REPO_URL: ${{ github.event.pull_request.head.repo.clone_url \|\| 'https://github.com/ROCm/ATOM.git' }}
		GITHUB_COMMIT_SHA: ${{ github.event.pull_request.head.sha \|\| github.event.head_commit.id }}



		docker run -dt --device=/dev/kfd $DEVICE_FLAG \
		docker run -dt --pull always --device=/dev/kfd $DEVICE_FLAG \

		RUNS=$(curl -s -H "$AUTH_HEADER" \
		"$API_URL/repos/ROCm/aiter/actions/workflows/$AITER_TEST_WORKFLOW_ID/runs?per_page=100&branch=main&event=push")

		echo '=== Installing amd-aiter from wheel ==='
		pip install /tmp/$WHL_NAME

Conversation

gyohuangxin commented Mar 11, 2026

Summary

Test plan

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview