sync

.github/scripts/extract_benchmark_results.py

@@ -86,36 +86,6 @@ def parse_args() -> Any:
         action=ValidateDir,
         help="the directory to keep the benchmark results",
     )
-    parser.add_argument(
-        "--repo",
-        type=str,
-        required=True,
-        help="which GitHub repo this workflow run belongs to",
-    )
-    parser.add_argument(
-        "--head-branch",
-        type=str,
-        required=True,
-        help="the head branch that runs",
-    )
-    parser.add_argument(
-        "--workflow-name",
-        type=str,
-        required=True,
-        help="the name of the benchmark workflow",
-    )
-    parser.add_argument(
-        "--workflow-run-id",
-        type=int,
-        required=True,
-        help="the id of the benchmark workflow",
-    )
-    parser.add_argument(
-        "--workflow-run-attempt",
-        type=int,
-        required=True,
-        help="which retry of the workflow this is",
-    )
     parser.add_argument(
         "--benchmark-configs",
         type=str,
@@ -153,9 +123,10 @@ def extract_android_benchmark_results(
         # This is to handle the case where there is no benchmark results
         warning(f"Fail to load the benchmark results from {artifact_s3_url}")
         return []
+    return []
 
 
-def initialize_ios_metadata(test_name: str) -> Dict[str, any]:
+def initialize_ios_metadata(test_name: str) -> Dict[str, Any]:
     """
     Extract the benchmark metadata from the test name, for example:
         test_forward_llama2_pte_iOS_17_2_1_iPhone15_4
@@ -364,14 +335,7 @@ def transform(
     app_type: str,
     benchmark_results: List,
     benchmark_config: Dict[str, str],
-    repo: str,
-    head_branch: str,
-    workflow_name: str,
-    workflow_run_id: int,
-    workflow_run_attempt: int,
     job_name: str,
-    job_id: int,
-    schema_version: str,
 ) -> List:
     """
     Transform the benchmark results into the format writable into the benchmark database
@@ -381,87 +345,51 @@ def transform(
     for r in benchmark_results:
         r["deviceInfo"]["device"] = job_name
 
-    if schema_version == "v2":
-        # TODO (huydhn): Clean up this branch after ExecuTorch dashboard migrates to v3
-        return [
-            {
-                # GH-info to identify where the benchmark is run
-                "repo": repo,
-                "head_branch": head_branch,
-                "workflow_id": workflow_run_id,
-                "run_attempt": workflow_run_attempt,
-                "job_id": job_id,
-                # The model
-                "name": f"{r['benchmarkModel']['name']} {r['benchmarkModel'].get('backend', '')}".strip(),
-                "dtype": (
-                    r["benchmarkModel"]["quantization"]
-                    if r["benchmarkModel"]["quantization"]
-                    else "unknown"
-                ),
-                # The metric value
-                "metric": r["metric"],
-                "actual": r["actualValue"],
-                "target": r["targetValue"],
-                # The device
-                "device": r["deviceInfo"]["device"],
-                "arch": r["deviceInfo"].get("os", ""),
-                # Not used here, just set it to something unique here
-                "filename": workflow_name,
-                "test_name": app_type,
-                "runner": job_name,
-            }
-            for r in benchmark_results
-        ]
-    elif schema_version == "v3":
-        v3_benchmark_results = []
-        # From https://github.com/pytorch/pytorch/wiki/How-to-integrate-with-PyTorch-OSS-benchmark-database
-        return [
-            {
-                "benchmark": {
-                    "name": "ExecuTorch",
-                    "mode": "inference",
-                    "extra_info": {
-                        "app_type": app_type,
-                        # Just keep a copy of the benchmark config here
-                        "benchmark_config": json.dumps(benchmark_config),
-                    },
-                },
-                "model": {
-                    "name": benchmark_config.get("model", r["benchmarkModel"]["name"]),
-                    "type": "OSS model",
-                    "backend": benchmark_config.get(
-                        "config", r["benchmarkModel"].get("backend", "")
-                    ),
+    # From https://github.com/pytorch/pytorch/wiki/How-to-integrate-with-PyTorch-OSS-benchmark-database
+    return [
+        {
+            "benchmark": {
+                "name": "ExecuTorch",
+                "mode": "inference",
+                "extra_info": {
+                    "app_type": app_type,
+                    # Just keep a copy of the benchmark config here
+                    "benchmark_config": json.dumps(benchmark_config),
                 },
-                "metric": {
-                    "name": r["metric"],
-                    "benchmark_values": [r["actualValue"]],
-                    "target_value": r["targetValue"],
-                    "extra_info": {
-                        "method": r.get("method", ""),
-                    },
+            },
+            "model": {
+                "name": benchmark_config.get("model", r["benchmarkModel"]["name"]),
+                "type": "OSS model",
+                "backend": benchmark_config.get(
+                    "config", r["benchmarkModel"].get("backend", "")
+                ),
+            },
+            "metric": {
+                "name": r["metric"],
+                "benchmark_values": [r["actualValue"]],
+                "target_value": r["targetValue"],
+                "extra_info": {
+                    "method": r.get("method", ""),
                 },
-                "runners": [
-                    {
-                        "name": r["deviceInfo"]["device"],
-                        "type": r["deviceInfo"]["os"],
-                        "avail_mem_in_gb": r["deviceInfo"].get("availMem", ""),
-                        "total_mem_in_gb": r["deviceInfo"].get("totalMem", ""),
-                    }
-                ],
-            }
-            for r in benchmark_results
-        ]
+            },
+            "runners": [
+                {
+                    "name": r["deviceInfo"]["device"],
+                    "type": r["deviceInfo"]["os"],
+                    "avail_mem_in_gb": r["deviceInfo"].get("availMem", ""),
+                    "total_mem_in_gb": r["deviceInfo"].get("totalMem", ""),
+                }
+            ],
+        }
+        for r in benchmark_results
+    ]
 
 
 def main() -> None:
     args = parse_args()
 
     # Across all devices, keeping both schemas for now until ExecuTorch dashboard migrates to v3
-    all_benchmark_results = {
-        "v2": [],
-        "v3": [],
-    }
+    all_benchmark_results = []
     benchmark_config = {}
 
     with open(args.artifacts) as f:
@@ -482,7 +410,7 @@ def main() -> None:
                 benchmark_config = read_benchmark_config(
                     artifact_s3_url, args.benchmark_configs
                 )
-
+            benchmark_results = []
             if app_type == "ANDROID_APP":
                 benchmark_results = extract_android_benchmark_results(
                     job_name, artifact_type, artifact_s3_url
@@ -494,32 +422,17 @@ def main() -> None:
                 )
 
             if benchmark_results:
-                for schema in all_benchmark_results.keys():
-                    results = transform(
-                        app_type,
-                        benchmark_results,
-                        benchmark_config,
-                        args.repo,
-                        args.head_branch,
-                        args.workflow_name,
-                        args.workflow_run_id,
-                        args.workflow_run_attempt,
-                        job_name,
-                        extract_job_id(args.artifacts),
-                        schema,
-                    )
-                    all_benchmark_results[schema].extend(results)
-
-    for schema in all_benchmark_results.keys():
-        if not all_benchmark_results.get(schema):
-            continue
-
-        output_dir = os.path.join(args.output_dir, schema)
-        os.makedirs(output_dir, exist_ok=True)
+                results = transform(
+                    app_type, benchmark_results, benchmark_config, job_name
+                )
+                all_benchmark_results.extend(results)
 
+        # add v3 in case we have higher version of schema
+        output_dir = os.path.join(args.output_dir, "v3")
+        os.makedirs(output_dir, exist_ok=True)
         output_file = os.path.basename(args.artifacts)
         with open(f"{output_dir}/{output_file}", "w") as f:
-            json.dump(all_benchmark_results[schema], f)
+            json.dump(all_benchmark_results, f)
 
 
 if __name__ == "__main__":

.github/workflows/_android.yml

@@ -30,6 +30,7 @@ jobs:
 
         # Build LLM Demo for Android
         bash build/build_android_library.sh ${ARTIFACTS_DIR_NAME}
+        bash build/build_android_instrumentation.sh
 
   # Running Android emulator directly on the runner and not using Docker
   run-emulator:

.github/workflows/android-perf.yml

@@ -462,29 +462,14 @@ jobs:
             ${CONDA_RUN} python .github/scripts/extract_benchmark_results.py \
               --artifacts "${ARTIFACTS_BY_JOB}" \
               --output-dir benchmark-results \
-              --repo ${{ github.repository }} \
-              --head-branch ${{ github.head_ref || github.ref_name }} \
-              --workflow-name "${{ github.workflow }}" \
-              --workflow-run-id ${{ github.run_id }} \
-              --workflow-run-attempt ${{ github.run_attempt }} \
               --benchmark-configs benchmark-configs
           done
 
-          for SCHEMA in v2 v3; do
-            for BENCHMARK_RESULTS in benchmark-results/"${SCHEMA}"/*.json; do
-              cat "${BENCHMARK_RESULTS}"
-              echo
-            done
+          for BENCHMARK_RESULTS in benchmark-results/v3/*.json; do
+            cat "${BENCHMARK_RESULTS}"
+            echo
           done
 
-      # TODO (huydhn): Remove v2 schema once the benchmark dashboard finishes the migration
-      - name: Upload the benchmark results (v2)
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
-        with:
-          benchmark-results-dir: benchmark-results/v2
-          dry-run: false
-          schema-version: v2
-
       - name: Upload the benchmark results (v3)
         uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
         with:

.github/workflows/apple-perf.yml

@@ -521,29 +521,14 @@ jobs:
             ${CONDA_RUN} python .github/scripts/extract_benchmark_results.py \
               --artifacts "${ARTIFACTS_BY_JOB}" \
               --output-dir benchmark-results \
-              --repo ${{ github.repository }} \
-              --head-branch ${{ github.head_ref || github.ref_name }} \
-              --workflow-name "${{ github.workflow }}" \
-              --workflow-run-id ${{ github.run_id }} \
-              --workflow-run-attempt ${{ github.run_attempt }} \
               --benchmark-configs benchmark-configs
           done
 
-          for SCHEMA in v2 v3; do
-            for BENCHMARK_RESULTS in benchmark-results/"${SCHEMA}"/*.json; do
-              cat "${BENCHMARK_RESULTS}"
-              echo
-            done
+          for BENCHMARK_RESULTS in benchmark-results/v3/*.json; do
+            cat "${BENCHMARK_RESULTS}"
+            echo
           done
 
-      # TODO (huydhn): Remove v2 schema once the benchmark dashboard finishes the migration
-      - name: Upload the benchmark results (v2)
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
-        with:
-          benchmark-results-dir: benchmark-results/v2
-          dry-run: false
-          schema-version: v2
-
       - name: Upload the benchmark results (v3)
         uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
         with: