Migrating changes from inference pr2162 + compliance run code changes-automotive (#452)

anandhu-eng · web-flow · commit 9c070fd8d203 · 2025-06-08T15:47:42.000+01:00
* migrating changes from inference pr2162

* update workflow file path

* rename master to main

* add round parameter

* corrected the logic

* adr given empty dictionary as the default value

* add skip calibration check

* skip calibration.md file check

* code change for supporting compliance runs

* create dummy calibration.md file when generating submission

* supply preprocessed dataset env variables to higher level scripts
diff --git a/.github/workflows/test-mlperf-inference-submission-generation.yml b/.github/workflows/test-mlperf-inference-submission-generation.yml
@@ -2,11 +2,12 @@
 
 name: MLC based MLPerf Inference Submission Generation
 
+
 on:
   pull_request:
-    branches: [ "master", "dev" ]
+    branches: [ "main", "dev" ]
     paths:
-      - '.github/workflows/mlperf-submission.yml'
+      - '.github/workflows/test-mlperf-inference-submission-generation.yml'
       - '**'
       - '!**.md'
   workflow_call:
@@ -28,12 +29,44 @@ jobs:
         python-version: [ "3.12" ]
         division: ["closed", "open", "closed-open"]
         category: ["datacenter", "edge"]
+        round: ["4.1", "5.0"]
         case: ["closed"]
         action: ["run", "docker"]
         exclude:
           - os: macos-latest
           - os: windows-latest
           - category: "edge"
+          - round: "4.1"
+          - case: "closed-edge"
+        include:
+          - os: ubuntu-latest
+            python-version: "3.12"
+            division: "closed"
+            category: "edge"
+            round: "5.0"
+            case: "closed-edge"
+            action: "run"
+          - os: ubuntu-latest
+            python-version: "3.12"
+            division: "closed"
+            category: "edge"
+            round: "5.0"
+            case: "closed-edge"
+            action: "docker"
+          - os: ubuntu-latest
+            python-version: "3.12"
+            division: "open"
+            category: "edge"
+            round: "5.0"
+            case: "closed-edge"
+            action: "run"
+          - os: ubuntu-latest
+            python-version: "3.12"
+            division: "open"
+            category: "edge"
+            round: "5.0"
+            case: "closed-edge"
+            action: "docker"
 
     steps:
     - uses: actions/checkout@v3
@@ -55,8 +88,7 @@ jobs:
     - name: Pull repo where test cases are uploaded
       run: |
         git clone -b submission-generation-examples https://github.com/mlcommons/inference.git submission_generation_examples
-    - name: Run Submission Generation - ${{ matrix.case }} ${{ matrix.action }} ${{ matrix.category }} ${{ matrix.division }} 
-      continue-on-error: true
+    - name: Run Submission Generation - round-${{ matrix.round }}${{ matrix.case }} ${{ matrix.action }} ${{ matrix.category }} ${{ matrix.division }} 
       run: |
         if [ "${{ matrix.case }}" == "closed" ]; then
           description="Test submission - contains closed edge and datacenter"
@@ -65,10 +97,18 @@ jobs:
         fi
         # Dynamically set the log group to simulate a dynamic step name
         echo "::group::$description"
-        if [ -z "${{ inputs.repo-url }}" ]; then
-          mlc ${{ matrix.action }} script --tags=generate,inference,submission --adr.compiler.tags=gcc --version=v5.0 --clean --preprocess_submission=yes --submission_base_dir=mysubmissions --results_dir=$PWD/submission_generation_tests/${{ matrix.case }}/ --run-checker --submitter=MLCommons --tar=yes --division=${{ matrix.division }} --env.MLC_DETERMINE_MEMORY_CONFIGURATION=yes --quiet --adr.inference-src.tags=_repo.${{ inputs.repo-url }},_branch.${{ inputs.ref }}
-          mlc ${{ matrix.action }} script --tags=run,submission,checker --submitter_id_off=mysubmitter_id --tar=yes --submission_dir=mysubmissions/submissions --submission_tar_file=mysubmission.tar.gz --adr.inference-src.tags=_repo.${{ inputs.repo-url }},_branch.${{ inputs.ref }}
+
+        docker_tags=""
+        if [ "${{ matrix.action }}" == "docker" ] && [ -z "${{ inputs.repo-url }}" ]; then
+          docker_tags="--docker_mlc_repo=${{ github.event.pull_request.head.repo.html_url }} --docker_mlc_repo_branch=${{ github.event.pull_request.head.ref }}"
+        fi
+
+        if [ -n "${{ inputs.repo-url }}" ]; then
+          mlc ${{ matrix.action }} script --tags=generate,inference,submission --adr.compiler.tags=gcc --version=v${{ matrix.round }} --clean --preprocess_submission=yes --submission_base_dir=mysubmissions --results_dir=$PWD/submission_generation_examples/submission_round_${{ matrix.round }}/${{ matrix.case }}/ --run_checker --submitter=MLCommons --tar=yes --division=${{ matrix.division }} --env.MLC_DETERMINE_MEMORY_CONFIGURATION=yes --quiet --adr.inference-src.tags=_repo.${{ inputs.repo-url }},_branch.${{ inputs.ref }} --skip_calibration_check=yes $docker_tags
+          mlc ${{ matrix.action }} script --tags=run,inference,submission,checker --submitter_id_off=mysubmitter_id --tar=yes --submission_dir=mysubmissions/mlperf_submission --submission_tar_file=mysubmission.tar.gz --quiet --src_version=v${{ matrix.round }} --adr.inference-src.tags=_repo.${{ inputs.repo-url }},_branch.${{ inputs.ref }} --skip_calibration_check=yes $docker_tags
         else
-          mlc ${{ matrix.action }} script --tags=generate,inference,submission --adr.compiler.tags=gcc --version=v5.0 --clean --preprocess_submission=yes --submission_base_dir=mysubmissions --results_dir=$PWD/submission_generation_tests/${{ matrix.case }}/ --run-checker --submitter=MLCommons --tar=yes --division=${{ matrix.division }} --env.MLC_DETERMINE_MEMORY_CONFIGURATION=yes --quiet
-          mlc ${{ matrix.action }} script --tags=run,submission,checker --submitter_id_off=mysubmitter_id --tar=yes --submission_dir=mysubmissions/submissions --submission_tar_file=mysubmission.tar.gz
+          mlc ${{ matrix.action }} script --tags=generate,inference,submission --adr.compiler.tags=gcc --version=v${{ matrix.round }} --clean --preprocess_submission=yes --submission_base_dir=mysubmissions --results_dir=$PWD/submission_generation_examples/submission_round_${{ matrix.round }}/${{ matrix.case }}/ --run_checker --submitter=MLCommons --tar=yes --division=${{ matrix.division }} --env.MLC_DETERMINE_MEMORY_CONFIGURATION=yes --quiet --skip_calibration_check=yes $docker_tags
+          mlc ${{ matrix.action }} script --tags=run,inference,submission,checker --submitter_id_off=mysubmitter_id --tar=yes --submission_dir=mysubmissions/mlperf_submission --submission_tar_file=mysubmission.tar.gz --quiet --src_version=v${{ matrix.round }} --skip_calibration_check=yes $docker_tags
         fi
+
+        echo "::endgroup::"
diff --git a/automation/script/docker.py b/automation/script/docker.py
@@ -274,7 +274,7 @@ def docker_run(self_module, i):
         env.setdefault('MLC_DOCKER_CACHE', docker_cache)
 
     image_repo = i.get('docker_image_repo', '')
-    add_deps_recursive = i.get('add_deps_recursive')
+    add_deps_recursive = i.get('add_deps_recursive', {})
 
     # Ensure Docker is available
     r = self_module.action_object.access(
diff --git a/script/app-mlperf-automotive-mlcommons-python/meta.yaml b/script/app-mlperf-automotive-mlcommons-python/meta.yaml
@@ -76,6 +76,7 @@ env_key_mappings:
 new_env_keys:
   - MLC_MLPERF_*
   - MLC_DATASET_*
+  - MLC_PREPROCESSED_DATASET_*
   - MLC_HW_NAME
   - MLC_COGNATA_ACCURACY_DUMP_FILE
   - MLC_OUTPUT_PREDICTIONS_PATH
diff --git a/script/app-mlperf-automotive/customize.py b/script/app-mlperf-automotive/customize.py
@@ -59,6 +59,33 @@ def postprocess(i):
     model = env['MLC_MODEL']
     model_full_name = env.get('MLC_ML_MODEL_FULL_NAME', model)
 
+    if mode == "accuracy" or mode == "compliance" and env[
+            'MLC_MLPERF_LOADGEN_COMPLIANCE_TEST'] == "TEST01":
+
+        out_baseline_accuracy_string = f"""> {q}{os.path.join(output_dir, "accuracy", "baseline_accuracy.txt")}{q} """
+        out_compliance_accuracy_string = f"""> {q}{os.path.join(output_dir, "accuracy", "compliance_accuracy.txt")}{q} """
+
+        if model == "ssd":
+            accuracy_filename = "accuracy_cognata.py"
+            accuracy_file_path = os.path.join(
+                env['MLC_MLPERF_INFERENCE_SSD_RESNET50_PATH'], accuracy_filename)
+            dataset_args = f""" --dataset-path {env['MLC_PREPROCESSED_DATASET_COGNATA_PATH']} --config baseline_8MP_ss_scales_fm1_5x5_all """
+            accuracy_log_file_option_name = " --mlperf-accuracy-file "
+
+        if model == "bevformer":
+            accuracy_filename = "accuracy_nuscenes_cpu.py"
+            accuracy_file_path = os.path.join(
+                env['MLC_MLPERF_INFERENCE_BEVFORMER_PATH'], accuracy_filename)
+            dataset_args = f""" --nuscenes-dir {env['MLC_PREPROCESSED_DATASET_NUSCENES_ACC_CHECKER_MIN_FILES_PATH']} --config {os.path.join(env['MLC_MLPERF_INFERENCE_BEVFORMER_PATH'], "projects" + "configs" + "bevformer" + "bevformer_tiny.py")} """
+            accuracy_log_file_option_name = " --mlperf-accuracy-file "
+
+        if model == "deeplabv3plus":
+            accuracy_filename = "accuracy_cognata.py"
+            accuracy_file_path = os.path.join(
+                env['MLC_MLPERF_INFERENCE_DEEPLABV3PLUS_PATH'], accuracy_filename)
+            dataset_args = f""" --dataset-path {env['MLC_PREPROCESSED_DATASET_COGNATA_PATH']} """
+            accuracy_log_file_option_name = " --mlperf-accuracy-file "
+
     scenario = env['MLC_MLPERF_LOADGEN_SCENARIO']
 
     if not os.path.exists(output_dir) or not os.path.exists(
@@ -357,6 +384,92 @@ def postprocess(i):
             with open("README-extra.md", "w") as fp:
                 fp.write(extra_readme)
 
+    elif mode == "compliance":
+        test = env.get("MLC_MLPERF_LOADGEN_COMPLIANCE_TEST", "TEST01")
+
+        RESULT_DIR = os.path.split(output_dir)[0]
+        COMPLIANCE_DIR = output_dir
+        OUTPUT_DIR = os.path.dirname(COMPLIANCE_DIR)
+
+        SCRIPT_PATH = os.path.join(
+            env['MLC_MLPERF_INFERENCE_SOURCE'],
+            "compliance",
+            test,
+            "run_verification.py")
+        if test == "TEST06":
+            cmd = f"""{env['MLC_PYTHON_BIN_WITH_PATH']}  {q}{SCRIPT_PATH}{q}  -c  {q}{COMPLIANCE_DIR}{q}  -o  {q}{OUTPUT_DIR}{q} --scenario {scenario} --dtype int32"""
+        else:
+            cmd = f"""{env['MLC_PYTHON_BIN_WITH_PATH']}  {q}{SCRIPT_PATH}{q}  -r {q}{RESULT_DIR}{q} -c  {q}{COMPLIANCE_DIR}{q}  -o  {q}{OUTPUT_DIR}{q}"""
+
+        logger.info(cmd)
+        os.system(cmd)
+
+        if test == "TEST01":
+
+            run_script_input = i['run_script_input']
+            automation = i['automation']
+
+            SCRIPT_PATH = os.path.join(env['MLC_MLPERF_INFERENCE_SOURCE'], "compliance", test,
+                                       "create_accuracy_baseline.sh")
+            TEST01_DIR = os.path.join(OUTPUT_DIR, "TEST01")
+            OUTPUT_DIR = os.path.join(OUTPUT_DIR, "TEST01", "accuracy")
+            if not os.path.exists(OUTPUT_DIR):
+                os.makedirs(OUTPUT_DIR)
+
+            ACCURACY_DIR = os.path.join(RESULT_DIR, "accuracy")
+            if not os.path.exists(ACCURACY_DIR):
+                logger.warning("Accuracy run not yet completed")
+                return {
+                    'return': 1, 'error': 'TEST01 needs accuracy run to be completed first'}
+
+            cmd = f"""cd {q}{TEST01_DIR}{q} &&  bash {q}{SCRIPT_PATH}{q} {q}{os.path.join(ACCURACY_DIR, "mlperf_log_accuracy.json")}{q} {q}{os.path.join(COMPLIANCE_DIR, "mlperf_log_accuracy.json")}{q} """
+            env['CMD'] = cmd
+            logger.info(cmd)
+            r = automation.run_native_script(
+                {'run_script_input': run_script_input, 'env': env, 'script_name': 'verify_accuracy'})
+            if r['return'] > 0:
+                return r
+
+            verify_accuracy_file = os.path.join(
+                TEST01_DIR, "verify_accuracy.txt")
+            with open(verify_accuracy_file, 'r') as file:
+                data = file.read().replace('\n', '\t')
+
+            if 'TEST PASS' not in data:
+                logger.warning(
+                    "\nDeterministic TEST01 failed... Trying with non-determinism.\n")
+            # #Normal test failed, trying the check with non-determinism
+
+                baseline_accuracy_file = os.path.join(
+                    TEST01_DIR, "mlperf_log_accuracy_baseline.json")
+                CMD = f"""cd {q}{ACCURACY_DIR}{q} && {q}{env['MLC_PYTHON_BIN_WITH_PATH']}{q} {q}{accuracy_filepath}{q} \
+{accuracy_log_file_option_name} {q}{baseline_accuracy_file}{q} {dataset_args} {datatype_option} {out_baseline_accuracy_string} """
+
+                env['CMD'] = CMD
+                r = automation.run_native_script(
+                    {'run_script_input': run_script_input, 'env': env, 'script_name': 'verify_accuracy'})
+                if r['return'] > 0:
+                    return r
+
+                if os.stat(baseline_accuracy_file).st_size == 0:
+                    return {'return': 1,
+                            'error': f"{baseline_accuracy_file} is empty"}
+
+                CMD = f"""cd {q}{ACCURACY_DIR}{q} && {q}{env['MLC_PYTHON_BIN_WITH_PATH']}{q} {q}{accuracy_filepath}{q} \
+{accuracy_log_file_option_name} {q}{os.path.join(TEST01_DIR, "mlperf_log_accuracy.json")}{q} {dataset_args} {datatype_option} \
+{out_compliance_accuracy_string} """
+
+                env['CMD'] = CMD
+                r = automation.run_native_script(
+                    {'run_script_input': run_script_input, 'env': env, 'script_name': 'verify_accuracy'})
+                if r['return'] > 0:
+                    return r
+        import submission_checker as checker
+        is_valid = checker.check_compliance_perf_dir(
+            COMPLIANCE_DIR) if test != "TEST06" else True
+        state['mlc-mlperf-inference-results'][state['MLC_SUT_CONFIG_NAME']
+                                              ][model][scenario][test] = "passed" if is_valid else "failed"
+
     if state.get(
             'abtf-inference-implementation') and state['abtf-inference-implementation'].get('version_info'):
         env['MLC_MLPERF_RUN_JSON_VERSION_INFO_FILE'] = os.path.join(
diff --git a/script/generate-mlperf-inference-submission/customize.py b/script/generate-mlperf-inference-submission/customize.py
@@ -166,6 +166,21 @@ def generate_submission(env, state, inp, submission_division, logger):
     if not os.path.isdir(path_submission):
         os.makedirs(path_submission)
 
+    # Save empty calibration.md file in the root directory and make it
+    # available for the submitters to fill
+    try:
+        calibration_readme_path = os.path.join(
+            path_submission, "calibration.md")
+        with open(calibration_readme_path, "w") as fp:
+            fp.write("MLPerf Inference Calibration and Quantization Details\n")
+        logger.info(
+            f"Created calibration.md file at {calibration_readme_path}")
+    except Exception as e:
+        logger.error(f"Error creating calibration.md file: {e}")
+        return {'return': 1, 'error': f"Error creating calibration.md file: {e}"}
+
+    logger.info(
+        f"Created calibration.md file at {calibration_readme_path}")
     # SUT base
     system = env.get('MLC_HW_NAME', 'default').replace(' ', '_')
 
diff --git a/script/generate-mlperf-inference-user-conf/customize.py b/script/generate-mlperf-inference-user-conf/customize.py
@@ -265,12 +265,20 @@ def preprocess(i):
         else:
             audit_path = test
 
-        audit_full_path = os.path.join(
-            env['MLC_MLPERF_INFERENCE_SOURCE'],
-            "compliance",
-            "nvidia",
-            audit_path,
-            "audit.config")
+        if env['MLC_BENCHMARK_GROUP'] == "automotive":
+            audit_full_path = os.path.join(
+                env['MLC_MLPERF_INFERENCE_SOURCE'],
+                "compliance",
+                audit_path,
+                "audit.config")
+        else:
+            audit_full_path = os.path.join(
+                env['MLC_MLPERF_INFERENCE_SOURCE'],
+                "compliance",
+                "nvidia",
+                audit_path,
+                "audit.config")
+
         env['MLC_MLPERF_INFERENCE_AUDIT_PATH'] = audit_full_path
         # copy the audit conf to the run directory incase the implementation is
         # not supporting the audit-conf path
@@ -518,12 +526,20 @@ def run_files_exist(mode, OUTPUT_DIR, run_files, env, logger):
 
         test = env['MLC_MLPERF_LOADGEN_COMPLIANCE_TEST']
 
-        SCRIPT_PATH = os.path.join(
-            env['MLC_MLPERF_INFERENCE_SOURCE'],
-            "compliance",
-            "nvidia",
-            test,
-            "run_verification.py")
+        if env['MLC_BENCHMARK_GROUP'] == "automotive":
+            SCRIPT_PATH = os.path.join(
+                env['MLC_MLPERF_INFERENCE_SOURCE'],
+                "compliance",
+                test,
+                "run_verification.py")
+        else:
+            SCRIPT_PATH = os.path.join(
+                env['MLC_MLPERF_INFERENCE_SOURCE'],
+                "compliance",
+                "nvidia",
+                test,
+                "run_verification.py")
+
         if test == "TEST06":
             cmd = f"{env['MLC_PYTHON_BIN_WITH_PATH']}  {SCRIPT_PATH}  -c  {COMPLIANCE_DIR}  -o  {OUTPUT_DIR} --scenario {scenario} --dtype int32"
         else:
diff --git a/script/get-ml-model-bevformer/meta.yaml b/script/get-ml-model-bevformer/meta.yaml
@@ -25,7 +25,7 @@ variations:
     env:
       MLC_MODEL_FORMAT: pth
       MLC_MODEL_RCLONE_FILEPATH: model_checkpoint_bevformer/bevformer_tiny_epoch_24.pth
-      MLC_ML_MODEL_FILENAME: bevformer_tiny_epoch_24.onnx
+      MLC_ML_MODEL_FILENAME: bevformer_tiny_epoch_24.pth
   mlc:
     group: download-src
     default: true
diff --git a/script/run-mlperf-automotive-app/customize.py b/script/run-mlperf-automotive-app/customize.py
@@ -116,6 +116,12 @@ def preprocess(i):
 
     test_list = []
 
+    # Add the compliance tests required for the models
+    if env['MLC_MODEL'] in ['ssd', 'bevformer', 'deeplabv3plus']:
+        test_list.append('TEST01')
+    if env['MLC_MODEL'] in ['bevformer', 'deeplabv3plus']:
+        test_list.append('TEST04')
+
     variation_benchmark_version = "_" + env["MLC_MLPERF_INFERENCE_VERSION"]
     variation_implementation = ",_" + \
         env.get("MLC_MLPERF_IMPLEMENTATION", "reference")
diff --git a/script/run-mlperf-inference-submission-checker/customize.py b/script/run-mlperf-inference-submission-checker/customize.py
@@ -58,6 +58,11 @@ def preprocess(i):
     else:
         power_check = ""
 
+    if is_true(env.get('MLC_MLPERF_SKIP_CALIBRATION_CHECK', 'no')):
+        calibration_check = " --skip-calibration-check"
+    else:
+        calibration_check = ""
+
     extra_args = ' ' + env.get('MLC_MLPERF_SUBMISSION_CHECKER_EXTRA_ARGS', '')
 
     x_submitter = ' --submitter ' + q + submitter + q if submitter != '' else ''
diff --git a/script/run-mlperf-inference-submission-checker/meta.yaml b/script/run-mlperf-inference-submission-checker/meta.yaml
@@ -86,6 +86,7 @@ input_mapping:
   repo_owner: MLC_MLPERF_RESULTS_GIT_REPO_OWNER
   skip_compliance: MLC_MLPERF_SKIP_COMPLIANCE
   skip_power_check: MLC_MLPERF_SKIP_POWER_CHECK
+  skip_calibration_check: MLC_MLPERF_SKIP_CALIBRATION_CHECK
   src_version: MLC_MLPERF_SUBMISSION_CHECKER_VERSION
   submission_dir: MLC_MLPERF_INFERENCE_SUBMISSION_DIR
   submitter: MLC_MLPERF_SUBMITTER
