Merge pull request #39 from arjunsuresh/mlperf-inference

Support get, docker

automation/script/module_misc.py

@@ -1721,6 +1721,13 @@ def docker(i):
     if image_repo == '':
         image_repo = 'cknowledge'
 
+    # Host system needs to have docker
+    r = self_module.cmind.access({'action':'run',
+                                'automation':'script',
+                                'tags': "get,docker"})
+    if r['return'] > 0:
+        return r
+
     for artifact in sorted(lst, key = lambda x: x.meta.get('alias','')):
 
         meta = artifact.meta

script/app-mlperf-inference-mlcommons-python/_cm.yaml

@@ -31,6 +31,9 @@ default_env:
   CM_MLPERF_SUT_NAME_IMPLEMENTATION_PREFIX: reference
   CM_MLPERF_SUT_NAME_RUN_CONFIG_SUFFIX: ''
 
+docker:
+  real_run: False
+
 # Map script inputs to environment variables
 input_mapping:
   count: CM_MLPERF_LOADGEN_QUERY_COUNT

script/app-mlperf-inference/_cm.yaml

@@ -319,8 +319,6 @@ variations:
       real_run: false
       run: true
       docker_input_mapping:
-        imagenet_path: IMAGENET_PATH
-        gptj_checkpoint_path: GPTJ_CHECKPOINT_PATH
         criteo_preprocessed_path: CRITEO_PREPROCESSED_PATH
         dlrm_data_path: DLRM_DATA_PATH
         intel_gptj_int8_model_path: CM_MLPERF_INFERENCE_INTEL_GPTJ_INT8_MODEL_PATH
@@ -904,9 +902,12 @@ variations:
   cuda,reference:
     docker:
       base_image: nvcr.io/nvidia/pytorch:24.03-py3
+
   cuda:
     docker:
       all_gpus: 'yes'
+      deps:
+        - tags: get,nvidia-docker 
     group:
       device
     env:

script/get-docker/customize.py

@@ -0,0 +1,64 @@
+from cmind import utils
+import os
+
+def preprocess(i):
+
+    os_info = i['os_info']
+
+    env = i['env']
+
+    automation = i['automation']
+
+    recursion_spaces = i['recursion_spaces']
+
+    file_name = 'docker.exe' if os_info['platform'] == 'windows' else 'docker'
+    env['FILE_NAME'] = file_name
+
+    if 'CM_DOCKER_BIN_WITH_PATH' not in env:
+        r = i['automation'].find_artifact({'file_name': file_name,
+                                       'env': env,
+                                       'os_info':os_info,
+                                       'default_path_env_key': 'PATH',
+                                       'detect_version':True,
+                                       'env_path_key':'CM_DOCKER_BIN_WITH_PATH',
+                                       'run_script_input':i['run_script_input'],
+                                       'recursion_spaces':recursion_spaces})
+        if r['return'] >0 : 
+            if r['return'] == 16:
+                run_file_name = "install"
+                r = automation.run_native_script({'run_script_input':i['run_script_input'], 'env':env, 'script_name':run_file_name})
+                if r['return'] >0: return r
+            else:
+                return r
+
+    return {'return':0}
+
+def detect_version(i):
+    r = i['automation'].parse_version({'match_text': r'Docker version\s*([\d.]+)',
+                                       'group_number': 1,
+                                       'env_key':'CM_DOCKER_VERSION',
+                                       'which_env':i['env']})
+    if r['return'] >0: return r
+
+    version = r['version']
+
+    print (i['recursion_spaces'] + '    Detected version: {}'.format(version))
+    return {'return':0, 'version':version}
+
+def postprocess(i):
+    env = i['env']
+
+    r = detect_version(i)
+
+    if r['return'] >0: return r
+
+    version = r['version']
+    found_file_path = env['CM_DOCKER_BIN_WITH_PATH']
+
+    found_path = os.path.dirname(found_file_path)
+    env['CM_DOCKER_INSTALLED_PATH'] = found_path
+    env['+PATH'] = [ found_path ]
+
+    env['CM_DOCKER_CACHE_TAGS'] = 'version-'+version
+
+    return {'return':0, 'version': version}

script/get-docker/install-centos.sh

@@ -0,0 +1,13 @@
+sudo yum install -y yum-utils
+sudo yum-config-manager --add-repo https://download.docker.com/linux/centos/docker-ce.repo
+sudo yum install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
+
+cmd="sudo usermod -aG docker $USER"
+echo "$cmd"
+eval "$cmd"
+test $? -eq 0 || exit $?
+
+echo "Please relogin to the shell so that the new group is  effective"
+exit 1
+#exec newgrp docker
+#sudo su - $USER

script/get-docker/run-ubuntu.sh → script/get-docker/install-ubuntu.sh

@@ -35,6 +35,9 @@ cmd="sudo usermod -aG docker $USER"
 echo "$cmd"
 eval "$cmd"
 test $? -eq 0 || exit $?
+
+echo "Please relogin to the shell so that the new group is  effective"
+exit 1
 #exec newgrp docker
 #sudo su - $USER
 

script/get-docker/install.bat

@@ -0,0 +1,2 @@
+echo "Please install docker to continue"
+exit 1

script/get-docker/install.sh

@@ -0,0 +1,2 @@
+echo "Please install docker to continue"
+exit 1

script/get-docker/run.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+docker --version  > tmp-ver.out
+test $? -eq 0 || exit 1

script/run-mlperf-inference-mobilenet-models/_cm.json

@@ -65,11 +65,14 @@
         "CM_MLPERF_SUBMISSION_MODE": "yes"
       }
     },
-    "populate-readme": {
+    "performance-and-accuracy": {
       "group": "run-mode",
+      "default": "true",
       "env": {
         "CM_MLPERF_FIND_PERFORMANCE_MODE": "no",
-        "CM_MLPERF_POPULATE_README": "yes"
+        "CM_MLPERF_PERFORMANCE_MODE": "yes",
+        "CM_MLPERF_ACCURACY_MODE": "yes",
+        "CM_MLPERF_SUBMISSION_MODE": "no"
       }
     },
     "all-models": {
@@ -86,6 +89,24 @@
         "CM_MLPERF_RUN_MOBILENETS": "yes"
       }
     },
+    "mobilenet-v1": {
+      "group": "model-selection",
+      "env": {
+        "CM_MLPERF_RUN_MOBILENET_V1": "yes"
+      }
+    },
+    "mobilenet-v2": {
+      "group": "model-selection",
+      "env": {
+        "CM_MLPERF_RUN_MOBILENET_V2": "yes"
+      }
+    },
+    "mobilenet-v3": {
+      "group": "model-selection",
+      "env": {
+        "CM_MLPERF_RUN_MOBILENET_V3": "yes"
+      }
+    },
     "efficientnet": {
       "group": "model-selection",
       "env": {

script/run-mlperf-inference-mobilenet-models/customize.py

@@ -21,7 +21,7 @@ def preprocess(i):
     quiet = (env.get('CM_QUIET', False) == 'yes')
     verbose = (env.get('CM_VERBOSE', False) == 'yes')
 
-    models = {
+    models_all = {
         "mobilenet": {
             "v1": {
                 "multiplier": [ "multiplier-1.0", "multiplier-0.75", "multiplier-0.5", "multiplier-0.25" ],
@@ -47,6 +47,22 @@ def preprocess(i):
                 }
             }
         }
+
+    models = {}
+    if env.get('CM_MLPERF_RUN_MOBILENET_V1', '') == "yes":
+        models['mobilenet'] = {}
+        models['mobilenet']['v1'] = models_all['mobilenet']['v1']
+    elif env.get('CM_MLPERF_RUN_MOBILENET_V2', '') == "yes":
+        models['mobilenet'] = {}
+        models['mobilenet']['v2'] = models_all['mobilenet']['v2']
+    elif env.get('CM_MLPERF_RUN_MOBILENET_V3', '') == "yes":
+        models['mobilenet'] = {}
+        models['mobilenet']['v3'] = models_all['mobilenet']['v3']
+    elif env.get('CM_MLPERF_RUN_MOBILENETS', '') == "yes":
+        models['mobilenet'] = models_all['mobilenet']
+    elif env.get('CM_MLPERF_RUN_EFFICIENTNETS', '') == "yes":
+        models['efficientnet'] = models_all['efficientnet']
+
     variation_strings = {}
     for t1 in models:
         variation_strings[t1] = []
@@ -73,12 +89,12 @@ def preprocess(i):
                             variation_list.append("_"+k3)
                         variation_strings[t1].append(",".join(variation_list))
 
-    if env.get('CM_MLPERF_POPULATE_README','') == "yes":
-        var="_populate-readme"
-        execution_mode="valid"
-    elif env.get('CM_MLPERF_SUBMISSION_MODE','') == "yes":
+    if env.get('CM_MLPERF_SUBMISSION_MODE','') == "yes":
         var="_submission"
         execution_mode="valid"
+    elif env.get('CM_MLPERF_ACCURACY_MODE','') == "yes" and env.get('CM_MLPERF_PERFORMANCE_MODE','') == "yes":
+        var="_full,_performance-and-accuracy"
+        execution_mode="valid"
     elif env.get('CM_MLPERF_ACCURACY_MODE','') == "yes":
         var="_full,_accuracy-only"
         execution_mode="valid"
@@ -151,22 +167,26 @@ def preprocess(i):
                 if env.get('CM_MLPERF_INFERENCE_SUBMISSION_DIR', '') != '':
                     cm_input['submission_dir'] = env['CM_MLPERF_INFERENCE_SUBMISSION_DIR']
 
-                if env.get('CM_MLPERF_ACCURACY_MODE','') == "yes":
-                    cm_input['mode'] = 'accuracy'
-
-                if env.get('CM_MLPERF_PERFORMANCE_MODE','') == "yes":
-                    cm_input['mode'] = 'performance'
-
                 if env.get('CM_MLPERF_FIND_PERFORMANCE_MODE','') == "yes" and env.get('CM_MLPERF_NO_RERUN','') != 'yes':
                     cm_input['rerun'] = True
 
                 if env.get('CM_MLPERF_POWER','') == "yes":
                     cm_input['power'] = 'yes'
 
-                print(cm_input)
-                r = cmind.access(cm_input)
-                if r['return'] > 0:
-                    return r
+                if env.get('CM_MLPERF_ACCURACY_MODE','') == "yes":
+                    cm_input['mode'] = 'accuracy'
+                    print(cm_input)
+                    r = cmind.access(cm_input)
+                    if r['return'] > 0:
+                        return r
+
+                if env.get('CM_MLPERF_PERFORMANCE_MODE','') == "yes":
+                    cm_input['mode'] = 'performance'
+
+                    print(cm_input)
+                    r = cmind.access(cm_input)
+                    if r['return'] > 0:
+                        return r
 
                 if env.get('CM_TEST_ONE_RUN', '') == "yes":
                     return {'return':0}