Collated reports (#40080)

ivarflakstad · web-flow · commit ebceef343aba · 2025-08-13T14:48:15.000+02:00
* Add initial collated reports script and job definition

* provide commit hash for this run. Also use hash in generated artifact name. Json formatting

* tidy

* Add option to upload collated reports to hf hub

* Add glob pattern for test report folders

* Fix glob

* Use machine_type as path filter instead of glob. Include machine_type in collated report
diff --git a/.github/workflows/collated-reports.yml b/.github/workflows/collated-reports.yml
@@ -0,0 +1,49 @@
+name: CI collated reports
+
+on:
+  workflow_call:
+    inputs:
+      job:
+        required: true
+        type: string
+      report_repo_id:
+        required: true
+        type: string
+      machine_type:
+        required: true
+        type: string
+      gpu_name:
+        description: Name of the GPU used for the job. Its enough that the value contains the name of the GPU, e.g. "noise-h100-more-noise". Case insensitive.
+        required: true
+        type: string
+
+jobs:
+  collated_reports:
+    name: Collated reports
+    runs-on: ubuntu-22.04
+    if: always()
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/download-artifact@v4
+
+      - name: Collated reports
+        shell: bash
+        env:
+          ACCESS_REPO_INFO_TOKEN: ${{ secrets.ACCESS_REPO_INFO_TOKEN }}
+          CI_SHA: ${{ github.sha }}
+          TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN: ${{ secrets.TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN }}
+        run: |
+          pip install huggingface_hub
+          python3 utils/collated_reports.py                  \
+            --path /transformers/reports/                    \
+            --machine-type ${{ inputs.machine_type }}        \
+            --commit-hash ${{ env.CI_SHA }}                  \
+            --job ${{ inputs.job }}                          \
+            --report-repo-id ${{ inputs.report_repo_id }}    \
+            --gpu-name ${{ inputs.gpu_name }}
+
+      - name: Upload collated reports
+        uses: actions/upload-artifact@v4
+        with:
+          name: collated_reports_${{ env.CI_SHA }}.json
+          path: collated_reports_${{ env.CI_SHA }}.json
diff --git a/utils/collated_reports.py b/utils/collated_reports.py
@@ -0,0 +1,219 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import json
+import subprocess
+from dataclasses import dataclass
+from pathlib import Path
+
+
+DEFAULT_GPU_NAMES = ["mi300", "mi325", "mi355", "h100", "a10"]
+
+
+def simplify_gpu_name(gpu_name: str, simplified_names: list[str]) -> str:
+    matches = []
+    for simplified_name in simplified_names:
+        if simplified_name in gpu_name:
+            matches.append(simplified_name)
+    if len(matches) == 1:
+        return matches[0]
+    return gpu_name
+
+
+def parse_short_summary_line(line: str) -> tuple[str | None, int]:
+    if line.startswith("PASSED"):
+        return "passed", 1
+    if line.startswith("FAILED"):
+        return "failed", 1
+    if line.startswith("SKIPPED"):
+        line = line.split("[", maxsplit=1)[1]
+        line = line.split("]", maxsplit=1)[0]
+        return "skipped", int(line)
+    if line.startswith("ERROR"):
+        return "error", 1
+    return None, 0
+
+
+def validate_path(p: str) -> Path:
+    # Validate path and apply glob pattern if provided
+    path = Path(p)
+    assert path.is_dir(), f"Path {path} is not a directory"
+    return path
+
+
+def get_gpu_name(gpu_name: str | None) -> str:
+    # Get GPU name if available
+    if gpu_name is None:
+        try:
+            import torch
+
+            gpu_name = torch.cuda.get_device_name()
+        except Exception as e:
+            print(f"Failed to get GPU name with {e}")
+            gpu_name = "unknown"
+    else:
+        gpu_name = gpu_name.replace(" ", "_").lower()
+        gpu_name = simplify_gpu_name(gpu_name, DEFAULT_GPU_NAMES)
+
+    return gpu_name
+
+
+def get_commit_hash(commit_hash: str | None) -> str:
+    # Get commit hash if available
+    if commit_hash is None:
+        try:
+            commit_hash = subprocess.check_output(["git", "rev-parse", "HEAD"]).decode("utf-8").strip()
+        except Exception as e:
+            print(f"Failed to get commit hash with {e}")
+            commit_hash = "unknown"
+
+    return commit_hash[:7]
+
+
+@dataclass
+class Args:
+    path: Path
+    machine_type: str
+    gpu_name: str
+    commit_hash: str
+    job: str | None
+    report_repo_id: str | None
+
+
+def get_arguments(args: argparse.Namespace) -> Args:
+    path = validate_path(args.path)
+    machine_type = args.machine_type
+    gpu_name = get_gpu_name(args.gpu_name)
+    commit_hash = get_commit_hash(args.commit_hash)
+    job = args.job
+    report_repo_id = args.report_repo_id
+    return Args(path, machine_type, gpu_name, commit_hash, job, report_repo_id)
+
+
+def upload_collated_report(job: str, report_repo_id: str, filename: str):
+    # Alternatively we can check for the existence of the collated_reports file and upload in notification_service.py
+    import os
+
+    from get_previous_daily_ci import get_last_daily_ci_run
+    from huggingface_hub import HfApi
+
+    api = HfApi()
+
+    # if it is not a scheduled run, upload the reports to a subfolder under `report_repo_folder`
+    report_repo_subfolder = ""
+    if os.getenv("GITHUB_EVENT_NAME") != "schedule":
+        report_repo_subfolder = f"{os.getenv('GITHUB_RUN_NUMBER')}-{os.getenv('GITHUB_RUN_ID')}"
+        report_repo_subfolder = f"runs/{report_repo_subfolder}"
+
+    workflow_run = get_last_daily_ci_run(
+        token=os.environ["ACCESS_REPO_INFO_TOKEN"], workflow_run_id=os.getenv("GITHUB_RUN_ID")
+    )
+    workflow_run_created_time = workflow_run["created_at"]
+    report_repo_folder = workflow_run_created_time.split("T")[0]
+
+    if report_repo_subfolder:
+        report_repo_folder = f"{report_repo_folder}/{report_repo_subfolder}"
+
+    api.upload_file(
+        path_or_fileobj=f"ci_results_{job}/{filename}",
+        path_in_repo=f"{report_repo_folder}/ci_results_{job}/{filename}",
+        repo_id=report_repo_id,
+        repo_type="dataset",
+        token=os.getenv("TRANSFORMERS_CI_RESULTS_UPLOAD_TOKEN"),
+    )
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Post process models test reports.")
+    parser.add_argument("--path", "-p", help="Path to the reports folder")
+    parser.add_argument(
+        "--machine-type", "-m", help="Process single or multi GPU results", choices=["single-gpu", "multi-gpu"]
+    )
+    parser.add_argument("--gpu-name", "-g", help="GPU name", default=None)
+    parser.add_argument("--commit-hash", "-c", help="Commit hash", default=None)
+    parser.add_argument("--job", "-j", help="Optional job name required for uploading reports", default=None)
+    parser.add_argument(
+        "--report-repo-id", "-r", help="Optional report repository ID required for uploading reports", default=None
+    )
+    args = get_arguments(parser.parse_args())
+
+    # Initialize accumulators for collated report
+    total_status_count = {
+        "passed": 0,
+        "failed": 0,
+        "skipped": 0,
+        "error": 0,
+        None: 0,
+    }
+    collated_report_buffer = []
+
+    path = args.path
+    machine_type = args.machine_type
+    gpu_name = args.gpu_name
+    commit_hash = args.commit_hash
+    job = args.job
+    report_repo_id = args.report_repo_id
+
+    # Find the origin directory based on machine type
+    origin = path
+    for p in path.iterdir():
+        if machine_type in p.name:
+            origin = p
+            break
+
+    # Loop through model directories and create collated reports
+    for model_dir in sorted(origin.iterdir()):
+        # Create a new entry for the model
+        model_name = model_dir.name.removesuffix("_test_reports")
+        report = {"model": model_name, "results": []}
+        results = []
+
+        # Read short summary
+        with open(model_dir / "summary_short.txt", "r") as f:
+            short_summary_lines = f.readlines()
+
+        # Parse short summary
+        for line in short_summary_lines[1:]:
+            status, count = parse_short_summary_line(line)
+            total_status_count[status] += count
+            if status:
+                result = {
+                    "status": status,
+                    "test": line.split(status.upper(), maxsplit=1)[1].strip(),
+                    "count": count,
+                }
+                results.append(result)
+
+        # Add short summaries to report
+        report["results"] = results
+
+        collated_report_buffer.append(report)
+
+    # Write collated report
+    with open(f"collated_reports_{commit_hash}.json", "w") as f:
+        json.dump(
+            {
+                "gpu_name": gpu_name,
+                "machine_type": machine_type,
+                "commit_hash": commit_hash,
+                "total_status_count": total_status_count,
+                "results": collated_report_buffer,
+            },
+            f,
+            indent=2,
+        )
+
+    if job and report_repo_id:
+        upload_collated_report(job, report_repo_id, f"collated_reports_{commit_hash}.json")
