[ Back to index ]

BERT-Large

BERT-99%: oBERT-Large Offline - DeepSparse

cm run script --tags=run,mlperf,inference,run-mlperf,_submission  \
   --adr.python.version_min=3.8 \
   --implementation=reference \
   --model=bert-99 \
   --precision=int8 \
   --backend=deepsparse \
   --device=cpu \
   --scenario=Offline \
   --mode=performance \
   --execution_mode=valid \
   --adr.mlperf-inference-implementation.max_batchsize=384 \
   --offline_target_qps=20 \
   --results_dir=$HOME/results_dir \
   --env.CM_MLPERF_NEURALMAGIC_MODEL_ZOO_STUB=zoo:nlp/question_answering/obert-large/pytorch/huggingface/squad/pruned95_quant-none-vnni

BERT-99%: MobileBERT Offline

cm run script --tags=run,mlperf,inference,run-mlperf,_submission  \
   --adr.python.name=mlperf \
   --adr.python.version_min=3.8 \
   --implementation=reference \
   --model=bert-99 \
   --precision=int8 \
   --backend=deepsparse \
   --device=cpu \
   --scenario=Offline \
   --mode=performance \
   --execution_mode=valid \
   --adr.mlperf-inference-implementation.max_batchsize=384 \
   --offline_target_qps=20 \
   --results_dir=$HOME/results_dir \
   --env.CM_MLPERF_NEURALMAGIC_MODEL_ZOO_STUB=zoo:nlp/question_answering/mobilebert-none/pytorch/huggingface/squad/14layer_pruned50_quant-none-vnni \
   --env.DEEPSPARSE_SEQLENS="64,128,192,256,384"

BERT-99.9%: MobileBERT Offline - DeepSparse

cm run script --tags=run,mlperf,inference,run-mlperf,_submission  \
   --adr.python.version_min=3.8 \
   --implementation=reference \
   --compliance=no \
   --model=bert-99 \
   --precision=int8 \
   --backend=deepsparse \
   --device=cpu \
   --scenario=Offline \
   --mode=performance \
   --execution_mode=valid \
   --adr.mlperf-inference-implementation.max_batchsize=384 \
   --offline_target_qps=20 \
   --results_dir=$HOME/results_dir \
   --env.DEEPSPARSE_SEQLENS="64,128,192,256,384" \
   --env.CM_MLPERF_NEURALMAGIC_MODEL_ZOO_STUB=zoo:nlp/question_answering/mobilebert-none/pytorch/huggingface/squad/base_quant-none

ResNet50

ResNet50 Offline - DeepSparse

cm run script --tags=run,mlperf,inference,run-mlperf,_submission  \
   --adr.python.version_min=3.8 \
   --implementation=reference \
   --model=resnet50 \
   --precision=int8 \
   --backend=deepsparse \
   --device=cpu \
   --scenario=Offline \
   --mode=performance \
   --execution_mode=valid \
   --adr.imagenet-preprocessed.tags=_pytorch \
   --adr.mlperf-inference-implementation.dataset=imagenet_pytorch \
   --adr.mlperf-inference-implementation.model=zoo:cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/pruned85_quant-none-vnni \
   --adr.mlperf-inference-implementation.max_batchsize=16 \
   --adr.mlperf-inference-implementation.num_threads=48 \
   --results_dir=$HOME/results_dir \
   --env.DEEPSPARSE_NUM_STREAMS=24 \
   --env.ENQUEUE_NUM_THREADS=2 \
   --offline_target_qps=204

Generate and upload MLPerf submission

Follow this guide to generate the submission tree and upload your results.

Questions? Suggestions?

Check the MLCommons Task Force on Automation and Reproducibility and get in touch via public Discord server.