Merge branch 'mlperf-inference' into docs

automation/script/module.py

@@ -975,7 +975,7 @@ def _run(self, i):
                 if str(state['docker'].get('run', True)).lower() in ['false', '0', 'no']:
                     print (recursion_spaces+'  - Skipping script::{} run as we are inside docker'.format(found_script_artifact))
                     return {'return': 0}
-                elif str(state['docker'].get('docker_real_run', True)).lower() in ['false', '0', 'no']:
+                elif str(state['docker'].get('real_run', True)).lower() in ['false', '0', 'no']:
                     print (recursion_spaces+'  - Doing fake run for script::{} as we are inside docker'.format(found_script_artifact))
                     fake_run = True
                     env['CM_TMP_FAKE_RUN']='yes'

script/add-custom-nvidia-system/_cm.yaml

@@ -6,7 +6,8 @@ automation_alias: script
 automation_uid: 5b4e0237da074764
 
 category: "MLPerf benchmark support"
-
+docker:
+  real_run: False
 
 # User-friendly tags to find this CM script
 tags:
@@ -94,6 +95,13 @@ variations:
     add_deps_recursive:
       nvidia-inference-common-code:
         tags: _ctuning
+  go:
+    group: code
+    add_deps_recursive:
+      nvidia-inference-common-code:
+        tags: _go
+
+
 
 
 versions:
@@ -111,3 +119,8 @@ versions:
     add_deps_recursive:
       nvidia-inference-common-code:
         version: r3.1
+
+  r4.0:
+    add_deps_recursive:
+      nvidia-inference-common-code:
+        version: r4.0

script/app-mlperf-inference-ctuning-cpp-tflite/_cm.json

@@ -118,6 +118,10 @@
       "tags": "get,ml-model,efficientnet,raw,_tflite"
     },
     {
+      "names": [
+        "tensorflow",
+	"tflite"
+      ],
       "tags": "get,tensorflow,lib,_tflite"
     },
     {

script/app-mlperf-inference-mlcommons-python/customize.py

@@ -177,7 +177,7 @@ def get_run_cmd_reference(os_info, env, scenario_extra_options, mode_extra_optio
 
         env['RUN_DIR'] = os.path.join(env['CM_MLPERF_INFERENCE_SOURCE'], "language", "gpt-j")
         cmd = env['CM_PYTHON_BIN_WITH_PATH'] +  \
-            " run.py --model-path=" + env['CM_ML_MODEL_FILE_WITH_PATH'] + ' --dataset-path=' + env['CM_DATASET_EVAL_PATH'] + " --scenario " + env['CM_MLPERF_LOADGEN_SCENARIO'] + " " + env['CM_MLPERF_LOADGEN_EXTRA_OPTIONS'] + \
+            " main.py --model-path=" + env['CM_ML_MODEL_FILE_WITH_PATH'] + ' --dataset-path=' + env['CM_DATASET_EVAL_PATH'] + " --scenario " + env['CM_MLPERF_LOADGEN_SCENARIO'] + " " + env['CM_MLPERF_LOADGEN_EXTRA_OPTIONS'] + \
             ' --dtype ' + env['CM_MLPERF_MODEL_PRECISION'] + \
             scenario_extra_options + mode_extra_options + dataset_options
         cmd = cmd.replace("--count", "--max_examples")

script/app-mlperf-inference-nvidia/_cm.yaml

@@ -261,7 +261,10 @@ deps:
       CM_MLPERF_NVIDIA_HARNESS_RUN_MODE:
       - run_harness
 
-  - tags: get,generic-python-lib,_package.nvmitten,_path./opt/nvmitten-0.1.3-cp38-cp38-linux_x86_64.whl
+  - tags: get,generic-python-lib,_package.nvmitten
+    update_tags_from_env_with_prefix:
+      _path.:
+        - CM_ENV_NVMITTEN_DOCKER_WHEEL_PATH
     enable_if_env:
       CM_RUN_STATE_DOCKER:
         - 'yes'
@@ -308,6 +311,23 @@ post_deps:
 
 # Variations to customize dependencies
 variations:
+  # MLPerf inference version
+  v4.0:
+    group: version
+    default: true
+    env:
+      CM_MLPERF_INFERENCE_VERSION: "v4.0"
+      CM_MLPERF_GPTJ_MODEL_FP8_PATH_SUFFIX: GPTJ-FP8-quantized
+    adr:
+      pytorch:
+        tags: _for-nvidia-mlperf-inference-v4.0
+  v3.1:
+    env:
+      CM_MLPERF_INFERENCE_VERSION: "v3.1"
+      CM_MLPERF_GPTJ_MODEL_FP8_PATH_SUFFIX: GPTJ-07142023.pth
+    adr:
+      pytorch:
+        tags: _for-nvidia-mlperf-inference-v3.1
   # Target devices
   cpu:
     group: device
@@ -366,6 +386,7 @@ variations:
     - tags: get,generic-python-lib,_transformers
     - tags: get,generic-python-lib,_safetensors
     - tags: get,generic-python-lib,_onnx
+    - tags: get,generic-python-lib,_onnx-graphsurgeon
 
   bert-99:
     group: model
@@ -476,18 +497,25 @@ variations:
     deps:
     - tags: get,generic-python-lib,_package.datasets
     - tags: get,generic-python-lib,_package.simplejson
+    - tags: get,generic-python-lib,_onnx
+    - tags: get,generic-python-lib,_transformers
+    - tags: get,generic-python-lib,_onnx-graphsurgeon
     env:
       CM_ML_MODEL_STARTING_WEIGHTS_FILENAME: "https://cloud.mlcommons.org/index.php/s/QAZ2oM94MkFtbQx/download"
 
   gptj_,build:
     deps:
-    - tags: install,pytorch,from.src,_for-nvidia-mlperf-inference-v3.1
+    - tags: install,pytorch,from.src
+      names:
+      - pytorch
     - tags: get,cmake
       version_min: "3.25.0"
 
   gptj_,build_engine:
     deps:
-    - tags: install,pytorch,from.src,_for-nvidia-mlperf-inference-v3.1
+    - tags: install,pytorch,from.src
+      names:
+      - pytorch
     - tags: get,cmake
       version_min: "3.25.0"
 
@@ -877,7 +905,9 @@ variations:
 
   gptj_,run_harness:
     deps:
-    - tags: install,pytorch,from.src,_for-nvidia-mlperf-inference-v3.1
+    - tags: install,pytorch,from.src
+      names:
+      - pytorch
     - tags: get,cmake
       version_min: "3.25.0"
     env:

script/app-mlperf-inference-nvidia/customize.py

@@ -10,6 +10,9 @@ def preprocess(i):
         return {'return':1, 'error': 'Windows is not supported in this script yet'}
     env = i['env']
 
+    if str(env.get('CM_RUN_STATE_DOCKER', '')).lower() in ['1', 'true', 'yes']:
+        return {'return': 0}
+
     if env.get('CM_MODEL', '') == '':
         return {'return': 1, 'error': 'Please select a variation specifying the model to run'}
 
@@ -153,7 +156,7 @@ def preprocess(i):
             cmds.append("make download_data BENCHMARKS='gptj'")
 
         fp32_model_path = os.path.join(env['MLPERF_SCRATCH_PATH'], 'models', 'GPTJ-6B', 'checkpoint-final')
-        fp8_model_path = os.path.join(env['MLPERF_SCRATCH_PATH'], 'models', 'GPTJ-6B', 'fp8-quantized-ammo', 'GPTJ-07142023.pth')
+        fp8_model_path = os.path.join(env['MLPERF_SCRATCH_PATH'], 'models', 'GPTJ-6B', 'fp8-quantized-ammo', env['CM_MLPERF_GPTJ_MODEL_FP8_PATH_SUFFIX'])
         vocab_path = os.path.join(env['MLPERF_SCRATCH_PATH'], 'models', 'bert', 'vocab.txt')
 
         if not os.path.exists(os.path.dirname(fp32_model_path)):
@@ -163,7 +166,8 @@ def preprocess(i):
 
         if not os.path.exists(fp32_model_path):
             env['CM_REQUIRE_GPTJ_MODEL_DOWNLOAD'] = 'yes' # download via prehook_deps
-            cmds.append(f"cp -r $CM_ML_MODEL_FILE_WITH_PATH {fp32_model_path}")
+            if make_command == "build_engine":
+                cmds.append(f"cp -r $CM_ML_MODEL_FILE_WITH_PATH {fp32_model_path}")
 
         model_name = "gptj"
         model_path = fp8_model_path

script/app-mlperf-inference/_cm.yaml

@@ -182,13 +182,16 @@ variations:
         tags: _float32
       librispeech-accuracy-script:
         tags: _int32
+      cnndm-accuracy-script:
+        tags: _int32
     env:
       CM_MLPERF_PYTHON: 'yes'
       CM_MLPERF_IMPLEMENTATION: mlcommons_python
       CM_SQUAD_ACCURACY_DTYPE: float32
       CM_IMAGENET_ACCURACY_DTYPE: float32
       CM_OPENIMAGES_ACCURACY_DTYPE: float32
       CM_LIBRISPEECH_ACCURACY_DTYPE: float32
+      CM_CNNDM_ACCURACY_DTYPE: int32
     prehook_deps:
       - names:
          - python-reference-mlperf-inference
@@ -235,6 +238,10 @@ variations:
     default_variations:
       backend: onnxruntime
 
+  nvidia-original,r4.1_default:
+    docker:
+      base_image: nvcr.io/nvidia/mlperf/mlperf-inference:mlpinf-v4.0-cuda12.2-cudnn8.9-x86_64-ubuntu20.04-public
+
   nvidia-original:
     docker:
       interactive: True
@@ -273,12 +280,6 @@ variations:
       CM_IMAGENET_ACCURACY_DTYPE: int32
       CM_CNNDM_ACCURACY_DTYPE: int32
       CM_LIBRISPEECH_ACCURACY_DTYPE: int8
-    deps:
-      - tags: get,cuda-devices
-        skip_if_env:
-          CM_CUDA_DEVICE_PROP_GLOBAL_MEMORY:
-            - "yes"
-            - "on"
     prehook_deps:
       - names:
          - nvidia-original-mlperf-inference
@@ -904,6 +905,12 @@ variations:
     add_deps_recursive:
       mlperf-inference-implementation:
         tags: _cuda
+    deps:
+      - tags: get,cuda-devices
+        skip_if_env:
+          CM_CUDA_DEVICE_PROP_GLOBAL_MEMORY:
+            - "yes"
+            - "on"
   rocm:
     docker:
       all_gpus: 'yes'
@@ -1142,6 +1149,25 @@ variations:
     default_env:
       CM_SKIP_SYS_UTILS: 'yes'
       CM_REGENERATE_MEASURE_FILES: 'yes'
+    env:
+      CM_ENV_NVMITTEN_DOCKER_WHEEL_PATH: '/opt/nvmitten-0.1.3-cp38-cp38-linux_x86_64.whl'
+
+  r4.1_default:
+    group:
+      reproducibility
+    add_deps_recursive:
+      nvidia-inference-common-code:
+        version: r4.0
+        tags: _go
+      nvidia-inference-server:
+        version: r4.0
+        tags: _go
+    default_env:
+      CM_SKIP_SYS_UTILS: 'yes'
+      CM_REGENERATE_MEASURE_FILES: 'yes'
+    env:
+      CM_ENV_NVMITTEN_DOCKER_WHEEL_PATH: '/opt/nvmitten-0.1.3b0-cp38-cp38-linux_x86_64.whl'
+
 
 invalid_variation_combinations:
   -
@@ -1240,6 +1266,7 @@ docker:
     - tags: get,mlperf,inference,results,dir
     - tags: get,mlperf,inference,submission,dir
   pre_run_cmds:
+    #- cm pull repo && cm run script --tags=get,git,repo,_repo.https://github.com/GATEOverflow/inference_results_v4.0.git --update
     - cm pull repo
   mounts:
    - "${{ CM_DATASET_IMAGENET_PATH }}:${{ CM_DATASET_IMAGENET_PATH }}"

script/authenticate-github-cli/_cm.yaml

@@ -0,0 +1,16 @@
+alias: authenticate-github-cli
+automation_alias: script
+automation_uid: 5b4e0237da074764
+cache: true
+input_mapping:
+  with_token: CM_GH_AUTH_TOKEN
+  with-token: CM_GH_AUTH_TOKEN
+tags:
+- auth
+- authenticate
+- github
+- gh
+- cli
+uid: 7b57673ac14a4337
+deps:
+  - tags: get,gh,cli

script/authenticate-github-cli/customize.py

@@ -0,0 +1,27 @@
+from cmind import utils
+import os
+
+def preprocess(i):
+
+    os_info = i['os_info']
+
+    env = i['env']
+
+    meta = i['meta']
+
+    automation = i['automation']
+
+    cmd = "gh auth login"
+    if env.get('CM_GH_AUTH_TOKEN', '') != '':
+        cmd = f" echo {env['CM_GH_AUTH_TOKEN']} | {cmd} --with-token"
+
+    env['CM_RUN_CMD'] = cmd
+    quiet = (env.get('CM_QUIET', False) == 'yes')
+
+    return {'return':0}
+
+def postprocess(i):
+
+    env = i['env']
+
+    return {'return':0}

script/authenticate-github-cli/run.bat

@@ -0,0 +1 @@
+rem native script

script/authenticate-github-cli/run.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+#CM Script location: ${CM_TMP_CURRENT_SCRIPT_PATH}
+
+#To export any variable
+#echo "VARIABLE_NAME=VARIABLE_VALUE" >>tmp-run-env.out
+
+#${CM_PYTHON_BIN_WITH_PATH} contains the path to python binary if "get,python" is added as a dependency
+
+echo "Running: "
+echo "${CM_RUN_CMD}"
+echo ""
+
+if [[ ${CM_FAKE_RUN} != "yes" ]]; then
+  eval "${CM_RUN_CMD}"
+  test $? -eq 0 || exit 1
+fi
+

script/build-mlperf-inference-server-nvidia/_cm.yaml

@@ -164,6 +164,11 @@ variations:
     add_deps_recursive:
       nvidia-inference-common-code:
         tags: _ctuning
+  go:
+    group: code
+    add_deps_recursive:
+      nvidia-inference-common-code:
+        tags: _go
   nvidia-only:
     group: code
     add_deps_recursive:
@@ -180,6 +185,22 @@ variations:
       nvidia-inference-common-code:
         tags: _mlcommons
 
+  r4.0:
+    group: version
+    add_deps_recursive:
+      nvidia-inference-common-code:
+        version: r4.0
+      nvidia-scratch-space:
+        tags: _version.4_1
+    deps:
+      - tags: install,pytorch,from.src,_for-nvidia-mlperf-inference-v4.0
+        names:
+        - pytorch
+        - torch
+      - tags: install,torchvision,from.src,_for-nvidia-mlperf-inference-v4.0
+        names:
+        - pytorchvision
+        - torchvision
 
 versions:
   r2.1:
@@ -213,6 +234,24 @@ versions:
         - torchvision
       - tags: install,nccl,libs,_cuda
 
+  r4.0:
+    add_deps_recursive:
+      nvidia-inference-common-code:
+        version: r4.0
+      nvidia-scratch-space:
+        tags: _version.4_1
+    env:
+      BUILD_TRTLLM: 1
+    deps:
+      - tags: install,pytorch,from.src,_for-nvidia-mlperf-inference-v4.0
+        names:
+        - pytorch
+        - torch
+      - tags: install,torchvision,from.src,_for-nvidia-mlperf-inference-v4.0
+        names:
+        - pytorchvision
+        - torchvision
+
 docker:
   skip_run_cmd: 'no'
   all_gpus: 'yes'

script/get-git-repo/_cm.json

@@ -17,6 +17,7 @@
     "folder": "CM_GIT_CHECKOUT_FOLDER",
     "patch": "CM_GIT_PATCH",
     "update": "CM_GIT_REPO_PULL",
+    "pull": "CM_GIT_REPO_PULL",
     "env_key": "CM_GIT_ENV_KEY",
     "submodules": "CM_GIT_RECURSE_SUBMODULES"
   },