MLPerf Inference Nvidia implementations #18
This file contains 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: "54 22 * * *" #to be adjusted | |
| jobs: | |
| run_nvidia: | |
| if: github.repository_owner == 'gateoverflow' | |
| runs-on: | |
| - self-hosted | |
| - linux | |
| - x64 | |
| - cuda | |
| - ${{ matrix.system }} | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| system: [ "GO-spr", "phoenix", "i9" ] | |
| 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 | |
| steps: | |
| - name: Test MLPerf Inference NVIDIA ${{ matrix.model }} | |
| run: | | |
| # Set hw_name based on matrix.system | |
| if [ "${{ matrix.system }}" = "GO-spr" ]; then | |
| hw_name="RTX4090x2" | |
| else | |
| hw_name="RTX4090x1" | |
| fi | |
| if [ -f "gh_action/bin/deactivate" ]; then source gh_action/bin/deactivate; fi | |
| python3 -m venv gh_action | |
| source gh_action/bin/activate | |
| export CM_REPOS=$HOME/GH_CM | |
| pip install --upgrade cm4mlops | |
| cm pull repo | |
| cm run script --tags=run-mlperf,inference,_all-scenarios,_submission,_full,_r4.1-dev --preprocess_submission=yes --adr.submission-checker-src.tags=_branch.dev --execution_mode=valid --gpu_name=rtx_4090 --pull_changes=yes --pull_inference_changes=yes --model=${{ matrix.model }} --submitter="MLCommons" --hw_name=$hw_name --implementation=nvidia --backend=tensorrt --category=datacenter,edge --division=closed --docker_dt=yes --docker_it=no --docker_cm_repo=gateoverflow@cm4mlops --adr.compiler.tags=gcc --device=cuda --use_dataset_from_host=yes --results_dir=$HOME/gh_action_results --submission_dir=$HOME/gh_action_submissions --clean --docker --quiet | |
| cm run script --tags=push,github,mlperf,inference,submission --repo_url=https://github.com/gateoverflow/mlperf_inference_unofficial_submissions_v5.0 --repo_branch=main --commit_message="Results from GH action on NVIDIA_$hw_name" --quiet --submission_dir=$HOME/gh_action_submissions --hw_name=$hw_name |