Fixes for nvidia-mlperf-inference (#152)

* Added variation for nvidia scratch space

* Fix docker_utils import in script module

Author: arjunsuresh

automation/script/docker_utils.py

@@ -104,7 +104,7 @@ def prepare_docker_inputs(input_params, docker_settings,
     keys = [
         "mlc_repo", "mlc_repo_branch", "base_image", "os", "os_version",
         "mlc_repos", "skip_mlc_sys_upgrade", "extra_sys_deps",
-        "gh_token", "fake_run_deps", "run_final_cmds", "real_run", "copy_files", "path"
+        "gh_token", "fake_run_deps", "run_final_cmds", "real_run", "copy_files", "path", "user"
     ]
 
     if run_stage:

automation/script/module.py

@@ -4950,7 +4950,7 @@ def find_cached_script(i):
                     # TODO Need to restrict the below check to within container
                     # env
                     i['tmp_dep_cached_path'] = dependent_cached_path
-                    import script.docker_utils
+                    from script import docker_utils
                     r = docker_utils.utils.get_container_path_script(i)
                     if not os.path.exists(r['value_env']):
                         # Need to rm this cache entry

script/app-mlperf-inference/meta.yaml

@@ -1671,6 +1671,8 @@ variations:
         tags: _ctuning
       intel-harness:
         tags: _v3.1
+      nvidia-scratch-space:
+        tags: _version.4_0-dev
     default_env:
       MLC_SKIP_SYS_UTILS: 'yes'
       MLC_REGENERATE_MEASURE_FILES: 'yes'
@@ -1709,6 +1711,8 @@ variations:
         tags: _mlcommons
       intel-harness:
         tags: _v4.0
+      nvidia-scratch-space:
+        tags: _version.4_1-dev
     default_env:
       MLC_SKIP_SYS_UTILS: 'yes'
       MLC_REGENERATE_MEASURE_FILES: 'yes'
@@ -1725,6 +1729,8 @@ variations:
         tags: _go
       intel-harness:
         tags: _v4.1
+      nvidia-scratch-space:
+        tags: _version.4_1
     default_env:
       MLC_SKIP_SYS_UTILS: 'yes'
       MLC_REGENERATE_MEASURE_FILES: 'yes'
@@ -1746,6 +1752,8 @@ variations:
         tags: _v4.1
       inference-src:
          version: r5.0
+      nvidia-scratch-space:
+        tags: _version.5.0-dev
     default_env:
       MLC_SKIP_SYS_UTILS: 'yes'
       MLC_REGENERATE_MEASURE_FILES: 'yes'
@@ -1901,6 +1909,7 @@ docker:
   interactive: True
   extra_run_args: ' --dns 8.8.8.8 --dns 8.8.4.4 --cap-add SYS_ADMIN  --cap-add SYS_TIME --security-opt apparmor=unconfined --security-opt seccomp=unconfined'
   os: ubuntu
+  user: mlcuser
   mlc_repo: mlcommons@mlperf-automations
   mlc_repo_branch: dev
   real_run: False

script/build-dockerfile/meta.yaml

@@ -56,6 +56,7 @@ input_mapping:
   skip_mlc_sys_upgrade: MLC_DOCKER_SKIP_MLC_SYS_UPGRADE
   push_image: MLC_DOCKER_PUSH_IMAGE
   docker_not_pull_update: MLC_DOCKER_NOT_PULL_UPDATE
+  user: MLC_DOCKER_USER
 
 new_env_keys:
 - MLC_DOCKERFILE_*

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

@@ -191,8 +191,6 @@ variations:
     add_deps_recursive:
       nvidia-inference-common-code:
         version: r4.0
-      nvidia-scratch-space:
-        tags: _version.4_1
     deps:
       - tags: get,generic,sys-util,_git-lfs
       - tags: install,pytorch,from.src,_for-nvidia-mlperf-inference-v4.0
@@ -344,6 +342,8 @@ docker:
     scratch_path: MLPERF_SCRATCH_PATH
   deps:
     - tags: get,mlperf,inference,nvidia,scratch,space
+      names:
+        - nvidia-scratch-space
     - tags: get,mlperf,inference,results,dir,local
     - tags: get,mlperf,inference,submission,dir,local
     - tags: get,nvidia-docker

script/get-mlperf-inference-nvidia-scratch-space/meta.yaml

@@ -39,8 +39,12 @@ variations:
       MLC_NVIDIA_SCRATCH_SPACE_VERSION: '4_1'
     group: version
   version.4_1-dev:
-    default: true
     env:
       MLC_NVIDIA_SCRATCH_SPACE_VERSION: 4_1-dev
     group: version
+  version.5_0-dev:
+    default: true
+    env:
+      MLC_NVIDIA_SCRATCH_SPACE_VERSION: 5_0-dev
+    group: version
 versions: {}