mlperf-language-processing.md

[ Back to index ]

Draft Stage

Run Commands

Quick submission run (short run)

cm run script --tags=run,mlperf,inference,generate-run-cmds,_submission,_short --submitter=name \
--lang=reference --model=bert-99 --backend=deepsparse --device=cpu --scenario=Offline  --quantized 
--results_dir=/home/cmuser/tmp2  --submission_dir=/home/cmuser/submission --clean

Customizations

1. _short -> _valid: For full submission run
2. --scenario -> one of SingleStream, Server for changing the scenarios
3. --backend -> one of onnxruntime, tf, pytorch, deepsparse
4. --quantized: only works for onnxruntime and deepsparse
5. --device -> one of cpu, cuda
6. --submitter: Name of the submitter excluding spaces
7. --clean: Clean everything and do not reuse previous run results