MLPerf Inference Nvidia implementations #1
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: MLPerf Inference Nvidia implementations | |
| on: | |
| schedule: | |
| - cron: "27 11 1 * *" | |
| jobs: | |
| run_nvidia: | |
| if: github.repository_owner == 'gateoverflow' | |
| timeout-minutes: 1440 | |
| runs-on: | |
| - self-hosted | |
| - linux | |
| - x64 | |
| - cuda | |
| - ${{ matrix.system }} | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| system: [ "GO-spr", "phoenix", "GO-i9" ] | |
| # system: [ "mlc-server" ] | |
| python-version: [ "3.12" ] | |
| model: [ "resnet50", "retinanet", "bert-99", "bert-99.9", "gptj-99.9", "3d-unet-99.9", "sdxl" ] | |
| exclude: | |
| - model: gptj-99.9 | |
| - system: phoenix1 | |
| - system: GO-i91 | |
| steps: | |
| - name: Test MLPerf Inference NVIDIA ${{ matrix.model }} | |
| env: | |
| gpu_name: rtx_4090 | |
| run: | | |
| # Set hw_name based on matrix.system | |
| if [ "${{ matrix.system }}" = "GO-spr" ]; then | |
| hw_name="RTX4090x2" | |
| gpu_name=rtx_4090 | |
| docker_string=" --docker --docker_recreate=yes" | |
| elif [ "${{ matrix.system }}" = "mlc-server" ]; then | |
| hw_name="H100x8" | |
| gpu_name=h100 | |
| docker_string=" " | |
| else | |
| hw_name="RTX4090x1" | |
| gpu_name=rtx_4090 | |
| docker_string=" --docker" | |
| fi | |
| if [ "${{ matrix.model }}" = "bert-99.9" ]; then | |
| submission_preprocessor_args=" --noinfer-low-accuracy-results" | |
| else | |
| submission_preprocessor_args="" | |
| fi | |
| category="datacenter,edge" | |
| if [ -f "gh_action/bin/deactivate" ]; then source gh_action/bin/deactivate; fi | |
| python3 -m venv gh_action | |
| source gh_action/bin/activate | |
| export MLC_REPOS=$HOME/GH_MLC | |
| pip install --upgrade mlcflow | |
| mlc pull repo mlcommons@mlperf-automations --branch=dev | |
| mlcr run-mlperf,inference,_all-scenarios,_submission,_full,_r5.0-dev --preprocess_submission=yes --pull_changes=yes --pull_inference_changes=yes --execution_mode=valid --gpu_name=$gpu_name --pull_changes=yes --pull_inference_changes=yes --model=${{ matrix.model }} --submitter="GATEOverflow" --hw_name=$hw_name --implementation=nvidia --backend=tensorrt --category=$category --division=closed --docker_dt --docker_mlc_repo=mlcommons@mlperf-automations --docker_mlc_repo_branch=dev --adr.compiler.tags=gcc --device=cuda --use_model_from_host=yes --use_dataset_from_host=yes --results_dir=$HOME/gh_action_results --submission_dir=$HOME/gh_action_submissions --clean $docker_string $submission_preprocessor_args --quiet | |
| #mlcr push,github,mlperf,inference,submission --repo_url=https://github.com/mlcommons/mlperf_inference_unofficial_submissions_v5.0 --repo_branch=auto-update --commit_message="Results from GH action on NVIDIA_$hw_name" --quiet --submission_dir=$HOME/gh_action_submissions --hw_name=$hw_name | |
| mlcr push,github,mlperf,inference,submission --repo_url=https://github.com/GATEOverflow/mlperf_inference_submissions_v5.0 --repo_branch=main --commit_message="Results from GH actions on NVIDIA_$hw_name" --quiet --submission_dir=$HOME/gh_action_submissions --hw_name=$hw_name | |