update

.github/runner-config.yml

@@ -2,12 +2,12 @@
 # Used by frameworks-devops-dashboard to display GPU architecture
 # Update this when runners are added/changed in CI workflows
 runners:
-  aiter-1gpu-runner:
-    gpu_arch: MI325
+  linux-aiter-mi300x-1:
+    gpu_arch: MI300X
     gpu_count: 1
 
-  aiter-8gpu-runner:
-    gpu_arch: MI325
+  linux-aiter-mi300x-8:
+    gpu_arch: MI300X
     gpu_count: 8
 
   linux-aiter-mi355-1:
@@ -17,3 +17,11 @@ runners:
   linux-aiter-mi355-8:
     gpu_arch: MI355
     gpu_count: 8
+
+  linux-aiter-mi35x-1:
+    gpu_arch: MI35X
+    gpu_count: 1
+
+  linux-aiter-mi35x-8:
+    gpu_arch: MI35X
+    gpu_count: 8

.github/scripts/check_signal.sh

@@ -1,34 +1,160 @@
 #!/bin/bash
 
-# This script attempts to download a pre-checks artifact from a GitHub workflow up to 5 times.
-# If the artifact is found and the signal indicates success, the workflow continues.
-# If the signal indicates failure, the workflow is skipped with details printed.
-# If the artifact cannot be downloaded after all retries, the workflow exits with an error.
-
-set -e
-
-ARTIFACT_NAME="checks-signal-${GITHUB_SHA:-${1}}"
-MAX_RETRIES=5
-
-for i in $(seq 1 $MAX_RETRIES); do
-  echo "Attempt $i: Downloading artifact..."
-  if gh run download --name "$ARTIFACT_NAME"; then
-    if [ -f checks_signal.txt ]; then
-      echo "Artifact $ARTIFACT_NAME downloaded successfully."
-      SIGNAL=$(head -n 1 checks_signal.txt)
-      if [ "$SIGNAL" = "success" ]; then
-        echo "Pre-checks passed, continuing workflow."
-        exit 0
-      else
-        echo "Pre-checks failed, skipping workflow. Details:"
-        tail -n +2 checks_signal.txt
-        exit 78  # 78 = neutral/skip
+# This script downloads the pre-checks artifact produced by the Checks workflow.
+# It scopes artifact lookup to the matching workflow run for the current branch and
+# head SHA, which avoids repo-wide artifact pagination and GitHub API rate limits.
+
+set -euo pipefail
+
+CHECKS_WORKFLOW_NAME="${CHECKS_WORKFLOW_NAME:-Checks}"
+CHECKS_SIGNAL_ARTIFACT_PREFIX="${CHECKS_SIGNAL_ARTIFACT_PREFIX:-checks-signal}"
+MAX_RETRIES="${MAX_RETRIES:-5}"
+RETRY_INTERVAL_SECONDS="${RETRY_INTERVAL_SECONDS:-30}"
+REPO="${GITHUB_REPOSITORY:-}"
+
+get_target_branch() {
+  if [ -n "${GITHUB_HEAD_REF:-}" ]; then
+    printf '%s\n' "${GITHUB_HEAD_REF}"
+    return
+  fi
+
+  if [ -n "${GITHUB_REF_NAME:-}" ]; then
+    printf '%s\n' "${GITHUB_REF_NAME}"
+    return
+  fi
+
+  python3 - <<'PY'
+import json
+import os
+
+event_path = os.environ.get("GITHUB_EVENT_PATH")
+if event_path and os.path.exists(event_path):
+    with open(event_path, encoding="utf-8") as fh:
+        data = json.load(fh)
+    print(data.get("pull_request", {}).get("head", {}).get("ref", ""))
+else:
+    print("")
+PY
+}
+
+get_target_head_sha() {
+  case "${GITHUB_EVENT_NAME:-}" in
+    pull_request|pull_request_target)
+      python3 - <<'PY'
+import json
+import os
+
+event_path = os.environ.get("GITHUB_EVENT_PATH")
+if event_path and os.path.exists(event_path):
+    with open(event_path, encoding="utf-8") as fh:
+        data = json.load(fh)
+    print(data.get("pull_request", {}).get("head", {}).get("sha", ""))
+else:
+    print("")
+PY
+      ;;
+    *)
+      printf '%s\n' "${GITHUB_SHA:-}"
+      ;;
+  esac
+}
+
+find_checks_run_id() {
+  local target_branch target_head_sha
+  target_branch="$(get_target_branch)"
+  target_head_sha="$(get_target_head_sha)"
+
+  if [ -z "${REPO}" ]; then
+    echo "GITHUB_REPOSITORY is required to locate the Checks workflow run." >&2
+    return 1
+  fi
+
+  if [ -z "${target_head_sha}" ]; then
+    echo "Could not determine the target head SHA for the Checks workflow run." >&2
+    return 1
+  fi
+
+  local -a gh_args=(
+    run list
+    --repo "${REPO}"
+    --workflow "${CHECKS_WORKFLOW_NAME}"
+    --limit 20
+    --json databaseId,headSha,headBranch,event,createdAt,status
+  )
+
+  if [ -n "${target_branch}" ]; then
+    gh_args+=(--branch "${target_branch}")
+  fi
+
+  # Nightly workflows reuse the Checks result from the push run on the same SHA.
+  if [ -n "${GITHUB_EVENT_NAME:-}" ] && [ "${GITHUB_EVENT_NAME}" != "schedule" ]; then
+    gh_args+=(--event "${GITHUB_EVENT_NAME}")
+  fi
+
+  gh "${gh_args[@]}" \
+    --jq "(map(select(.headSha == \"${target_head_sha}\")) | first | .databaseId) // empty"
+}
+
+find_signal_artifact_name() {
+  local run_id="$1"
+
+  gh api "repos/${REPO}/actions/runs/${run_id}/artifacts" | python3 -c '
+import json
+import sys
+
+prefix = sys.argv[1]
+data = json.load(sys.stdin)
+
+matching = sorted(
+    (
+        artifact
+        for artifact in data.get("artifacts", [])
+        if not artifact.get("expired")
+        and (
+            artifact.get("name") == prefix
+            or artifact.get("name", "").startswith(f"{prefix}-")
+        )
+    ),
+    key=lambda artifact: artifact.get("created_at", ""),
+)
+
+print(matching[-1]["name"] if matching else "")
+' "${CHECKS_SIGNAL_ARTIFACT_PREFIX}"
+}
+
+for i in $(seq 1 "${MAX_RETRIES}"); do
+  echo "Attempt ${i}: Locating ${CHECKS_WORKFLOW_NAME} workflow run..."
+  rm -f checks_signal.txt
+
+  RUN_ID="$(find_checks_run_id || true)"
+  if [ -n "${RUN_ID}" ]; then
+    ARTIFACT_NAME="$(find_signal_artifact_name "${RUN_ID}" || true)"
+    if [ -z "${ARTIFACT_NAME}" ]; then
+      echo "Attempt ${i}: No ${CHECKS_SIGNAL_ARTIFACT_PREFIX} artifact found in run ${RUN_ID} yet."
+    else
+      echo "Attempt ${i}: Downloading artifact '${ARTIFACT_NAME}' from run ${RUN_ID}..."
+      if gh run download "${RUN_ID}" --repo "${REPO}" --name "${ARTIFACT_NAME}"; then
+        if [ -f checks_signal.txt ]; then
+          echo "Artifact ${ARTIFACT_NAME} downloaded successfully."
+          SIGNAL="$(head -n 1 checks_signal.txt)"
+          if [ "${SIGNAL}" = "success" ]; then
+            echo "Pre-checks passed, continuing workflow."
+            exit 0
+          fi
+
+          echo "Pre-checks failed, skipping workflow. Details:"
+          tail -n +2 checks_signal.txt
+          exit 78  # 78 = neutral/skip
+        fi
       fi
     fi
+  else
+    echo "Attempt ${i}: Matching ${CHECKS_WORKFLOW_NAME} run not found yet."
   fi
-  echo "Artifact not found, retrying in 30s..."
-  sleep 30
+
+  echo "Artifact not ready yet, retrying in ${RETRY_INTERVAL_SECONDS}s..."
+  sleep "${RETRY_INTERVAL_SECONDS}"
 done
 
-echo "Failed to download pre-checks artifact after $MAX_RETRIES attempts. Exiting workflow."
+echo "Failed to download pre-checks artifact after ${MAX_RETRIES} attempts. Exiting workflow."
 exit 1

.github/scripts/generate_summary.py

@@ -0,0 +1,148 @@
+#!/usr/bin/env python3
+"""Generate GitHub Actions job summaries for Aiter CI.
+
+Usage:
+    python3 .github/scripts/generate_summary.py build
+    python3 .github/scripts/generate_summary.py promote
+
+Each mode reads its inputs from environment variables and appends
+Markdown to $GITHUB_STEP_SUMMARY.
+"""
+
+import os
+import sys
+from pathlib import Path
+
+DOMAIN_MAP = {
+    "nightlies": "rocm.frameworks-nightlies.amd.com",
+    "devreleases": "rocm.frameworks-devreleases.amd.com",
+    "prereleases": "rocm.frameworks-prereleases.amd.com",
+    "release": "rocm.frameworks.amd.com",
+}
+
+
+def _out(path: Path, line: str = "") -> None:
+    with open(path, "a") as f:
+        f.write(line + "\n")
+
+
+def _table(path: Path, headers: list[str], rows: list[list[str]]) -> None:
+    _out(path, "| " + " | ".join(headers) + " |")
+    _out(path, "| " + " | ".join("---" for _ in headers) + " |")
+    for row in rows:
+        _out(path, "| " + " | ".join(row) + " |")
+    _out(path)
+
+
+# ── Build summary ───────────────────────────────────────────────────────────
+
+
+def _get_index_url(release_type: str, gpu_archs: str = "gfx942-gfx950") -> str | None:
+    domain = DOMAIN_MAP.get(release_type)
+    if not domain:
+        return None
+    return f"https://{domain}/whl/{gpu_archs}/"
+
+
+def build_summary(summary: Path) -> None:
+    docker_image = os.environ.get("SUMMARY_DOCKER_IMAGE", "unknown")
+    python_version = os.environ.get("SUMMARY_PYTHON_VERSION", "unknown")
+    release_type = os.environ.get("SUMMARY_RELEASE_TYPE", "unknown")
+    gpu_archs = os.environ.get("SUMMARY_GPU_ARCHS", "unknown")
+    wheel_dir = os.environ.get("SUMMARY_WHEEL_DIR", "dist")
+    index_url = _get_index_url(release_type, gpu_archs.replace(";", "-"))
+
+    _out(summary, f"## Build Summary - Python {python_version}")
+    _out(summary)
+    rows = [
+        ["Python version", f"`{python_version}`"],
+        ["Docker image", f"`{docker_image}`"],
+        ["Release type", f"`{release_type}`"],
+        ["GPU architectures", f"`{gpu_archs}`"],
+    ]
+    if index_url:
+        rows.append(["Index URL", index_url])
+    _table(summary, ["Item", "Value"], rows)
+
+    _out(summary, "### Wheels")
+    _out(summary, "```")
+    whl_dir = Path(wheel_dir)
+    wheels = sorted(whl_dir.glob("*.whl")) if whl_dir.is_dir() else []
+    if wheels:
+        for w in wheels:
+            size_mb = w.stat().st_size / (1024 * 1024)
+            _out(summary, f"  {w.name}  ({size_mb:.1f} MB)")
+    else:
+        _out(summary, "  No wheels found")
+    _out(summary, "```")
+
+
+# ── Promote summary ─────────────────────────────────────────────────────────
+
+
+def promote_summary(summary: Path) -> None:
+    release_type = os.environ.get("SUMMARY_RELEASE_TYPE", "unknown")
+    s3_dest = os.environ.get("SUMMARY_S3_DEST", "")
+    wheel_names = os.environ.get("SUMMARY_WHEEL_NAMES", "").strip()
+
+    url_path = (
+        "/".join(s3_dest.split("/")[3:])
+        if s3_dest.startswith("s3://")
+        else "whl/gfx942-gfx950"
+    )
+    domain = DOMAIN_MAP.get(release_type)
+    index_url = f"https://{domain}/{url_path}/" if domain else None
+
+    _out(summary, "## Promote Summary")
+    _out(summary)
+    rows = [["Release type", f"`{release_type}`"]]
+    if index_url:
+        rows.append(["Index URL", index_url])
+    _table(summary, ["Item", "Value"], rows)
+
+    if wheel_names:
+        _out(summary, "### Promoted Wheels")
+        _out(summary, "```")
+        for whl in wheel_names.split():
+            _out(summary, f"  {whl}")
+        _out(summary, "```")
+        _out(summary)
+
+    if index_url:
+        _out(summary, "### Install Instructions")
+        _out(summary)
+        _out(summary, "**Using pip:**")
+        _out(summary, "```bash")
+        _out(summary, f"pip install --extra-index-url {index_url} amd-aiter")
+        _out(summary, "```")
+        _out(summary)
+        _out(summary, "**Using uv:**")
+        _out(summary, "```bash")
+        _out(summary, f"uv pip install --extra-index-url {index_url} amd-aiter")
+        _out(summary, "```")
+        _out(summary)
+
+
+# ── Main ────────────────────────────────────────────────────────────────────
+
+MODES = {
+    "build": build_summary,
+    "promote": promote_summary,
+}
+
+
+def main() -> None:
+    if len(sys.argv) != 2 or sys.argv[1] not in MODES:
+        print(f"Usage: {sys.argv[0]} {{{','.join(MODES)}}}", file=sys.stderr)
+        sys.exit(1)
+
+    summary_path = os.environ.get("GITHUB_STEP_SUMMARY")
+    if not summary_path:
+        print("GITHUB_STEP_SUMMARY is not set", file=sys.stderr)
+        sys.exit(1)
+
+    MODES[sys.argv[1]](Path(summary_path))
+
+
+if __name__ == "__main__":
+    main()