Merge pull request #588 from mlcommons/mlperf-inference

Sync main <- Mlperf inference for November 2024 release

.github/workflows/test-cm-based-submission-generation.yml

@@ -19,12 +19,29 @@ jobs:
         python-version: [ "3.12" ]
         division: ["closed", "open", "closed-open"]
         category: ["datacenter", "edge"]
-        case: ["closed", "closed-no-compliance", "closed-power", "closed-failed-power-logs", "case-3", "case-7", "case-8"]
+        case: ["closed", "closed-no-compliance", "closed-power", "closed-failed-power-logs", "case-1", "case-2", "case-3", "case-5", "case-6", "case-7", "case-8"]
         action: ["run", "docker"]
         exclude:
           - os: macos-latest
           - os: windows-latest
           - category: "edge"
+          - case: case-1
+            division: closed
+          - case: case-1
+            division: closed-open
+          - case: case-2
+            division: closed
+          - case: case-2
+            division: closed-open
+          - case: case-5
+            division: closed
+          - case: case-5
+            division: closed-open
+          - case: case-6
+            division: closed
+          - case: case-6
+            division: closed-open
+
     steps:
     - uses: actions/checkout@v4
     - name: Set up Python ${{ matrix.python-version }}
@@ -39,6 +56,7 @@ jobs:
       run: |
         git clone -b submission-generation-tests https://github.com/mlcommons/inference.git submission_generation_tests
     - name: Run Submission Generation - ${{ matrix.case }} ${{ matrix.action }} ${{ matrix.category }} ${{ matrix.division }} 
+      continue-on-error: true
       run: |
         if [ "${{ matrix.case }}" == "case-3" ]; then
           description="Submission generation (model_mapping.json not present but model name matches with official one)"
@@ -63,5 +81,18 @@ jobs:
         # Dynamically set the log group to simulate a dynamic step name
         echo "::group::$description"
         cm ${{ matrix.action }} script --tags=generate,inference,submission --adr.submission-checker-src.tags=_branch.dev --clean --preprocess_submission=yes --results_dir=$PWD/submission_generation_tests/${{ matrix.case }}/ --run-checker --submitter=MLCommons --tar=yes --env.CM_TAR_OUTFILE=submission.tar.gz --division=${{ matrix.division }} --env.CM_DETERMINE_MEMORY_CONFIGURATION=yes --quiet $extra_run_args
+        exit_status=$?
+        echo "Exit status for the job ${description} ${exit_status}"
+        if [[ "${{ matrix.case }}" == "case-5" || "${{ matrix.case }}" == "case-6" ]]; then
+          # For cases 5 and 6, exit status should be 0 if cm command fails, 1 if it succeeds
+          if [[ ${exit_status} -ne 0 ]]; then
+            exit 0
+          else
+            exit ${exit_status}
+          fi
+        else
+          # For other cases, exit with the original status
+          test ${exit_status} -eq 0 || exit ${exit_status}
+        fi
         echo "::endgroup::"
   

VERSION

@@ -1 +1 @@
-0.4.0
+0.4.1

script/generate-mlperf-inference-submission/_cm.yaml

@@ -83,6 +83,7 @@ input_mapping:
   submitter: CM_MLPERF_SUBMITTER
   sw_notes_extra: CM_MLPERF_SUT_SW_NOTES_EXTRA
   tar: CM_TAR_SUBMISSION_DIR
+  get_platform_details: CM_GET_PLATFORM_DETAILS
 post_deps:
 - enable_if_env:
     CM_RUN_MLPERF_ACCURACY:

script/generate-mlperf-inference-submission/customize.py

@@ -240,6 +240,7 @@ def generate_submission(env, state, inp, submission_division):
         compliance_path = os.path.join(path_submission, "compliance", sub_res)
         system_path = os.path.join(path_submission, "systems")
         submission_system_path = system_path
+
         if not os.path.isdir(submission_system_path):
             os.makedirs(submission_system_path)
         system_file = os.path.join(submission_system_path, sub_res+".json")
@@ -273,6 +274,8 @@ def generate_submission(env, state, inp, submission_division):
 
             print('* MLPerf inference model: {}'.format(model))
             for scenario in scenarios:
+                # the system_info.txt is copied from the mode directory if found, else it would be looked under scenario directory
+                system_info_from_mode = False
                 results[model][scenario] = {}
                 result_scenario_path = os.path.join(result_model_path, scenario)
                 submission_scenario_path = os.path.join(submission_model_path, scenario)
@@ -429,6 +432,8 @@ def generate_submission(env, state, inp, submission_division):
                             elif f in [ "README.md", "README-extra.md", "cm-version-info.json", "os_info.json", "cpu_info.json", "pip_freeze.json", "system_info.txt", "cm-deps.png", "cm-deps.mmd" ] and mode == "performance":
                                 shutil.copy(os.path.join(result_mode_path, f), os.path.join(submission_measurement_path, f))
                                 if f == "system_info.txt" and not platform_info_file:
+                                    # the first found system_info.txt will be taken as platform info file for a specific model to be placed in 
+                                    # measurements-model folder when generating the final submission 
                                     platform_info_file = os.path.join(result_mode_path, f)
                             elif f in [ "console.out" ]:
                                 shutil.copy(os.path.join(result_mode_path, f), os.path.join(submission_measurement_path, mode+"_"+f))
@@ -445,6 +450,9 @@ def generate_submission(env, state, inp, submission_division):
                         p_target = os.path.join(submission_results_path, f)
                         shutil.copy(os.path.join(result_mode_path, f), p_target)
 
+                if os.path.exists(os.path.join(result_scenario_path, "system_info.txt")):
+                    shutil.copy(os.path.join(result_scenario_path, "system_info.txt"), os.path.join(submission_measurement_path, f))
+                    platform_info_file = os.path.join(result_scenario_path, "system_info.txt")
 
                 readme_file = os.path.join(submission_measurement_path, "README.md")
                 if not os.path.exists(readme_file):
@@ -459,24 +467,39 @@ def generate_submission(env, state, inp, submission_division):
                 with open(readme_file, mode='a') as f:
                     f.write(result_string)
 
-            #Copy system_info.txt to the submission measurements model folder if any scenario performance run has it
+            # Copy system_info.txt to the submission measurements model folder if any scenario performance run has it
             sys_info_file = None
+
             if os.path.exists(os.path.join(result_model_path, "system_info.txt")):
                 sys_info_file = os.path.join(result_model_path, "system_info.txt")
             elif platform_info_file:
                 sys_info_file = platform_info_file
+
             if sys_info_file:
                 model_platform_info_file = sys_info_file
                 shutil.copy(sys_info_file, os.path.join(measurement_model_path, "system_info.txt"))
 
         #Copy system_info.txt to the submission measurements folder if any model performance run has it
         sys_info_file = None
+
         if os.path.exists(os.path.join(result_path, "system_info.txt")):
             sys_info_file = os.path.join(result_path, "system_info.txt")
         elif model_platform_info_file:
             sys_info_file = model_platform_info_file
+
         if sys_info_file:
             shutil.copy(sys_info_file, os.path.join(measurement_path, "system_info.txt"))
+        else:
+            if env.get('CM_GET_PLATFORM_DETAILS', '') == "yes":
+                cm_input = {'action': 'run',
+                            'automation': 'script',
+                            'tags': 'get,platform,details',
+                            'env': {'CM_PLATFORM_DETAILS_FILE_PATH': os.path.join(measurement_path, "system_info.txt")},
+                            'quiet': True
+                            }
+                r = cmind.access(cm_input)
+                if r['return'] > 0:
+                    return r
 
 
         with open(system_file, "w") as fp: