Neural Magic DeepSparse 1.5.0

pts/deepsparse-1.5.0 - 07 June 2023 - Update against Deepsparse 1.5 upstream.

install.sh

#!/bin/sh
rm -rf ~/.cache/
pip3 install --user deepsparse==1.5.0 sparsezoo==1.5.1
~/.local/bin/sparsezoo.download zoo:nlp/text_classification/bert-base/pytorch/huggingface/sst2/base-none
~/.local/bin/sparsezoo.download zoo:nlp/text_classification/distilbert-none/pytorch/huggingface/mnli/base-none
~/.local/bin/sparsezoo.download zoo:cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/base-none
~/.local/bin/sparsezoo.download zoo:nlp/token_classification/bert-base/pytorch/huggingface/conll2003/base-none
~/.local/bin/sparsezoo.download zoo:nlp/question_answering/bert-base/pytorch/huggingface/squad/12layer_pruned90-none
~/.local/bin/sparsezoo.download zoo:cv/detection/yolov5-s/pytorch/ultralytics/coco/base-none
~/.local/bin/sparsezoo.download zoo:nlp/document_classification/obert-base/pytorch/huggingface/imdb/base-none
~/.local/bin/sparsezoo.download zoo:nlp/sentiment_analysis/bert-base/pytorch/huggingface/sst2/12layer_pruned90-none
~/.local/bin/sparsezoo.download zoo:cv/segmentation/yolact-darknet53/pytorch/dbolya/coco/pruned90-none
echo $? > ~/install-exit-status
echo "#!/bin/sh
~/.local/bin/deepsparse.benchmark \$@ > \$LOG_FILE 2>&1
echo \$? > ~/test-exit-status" > deepsparse
chmod +x deepsparse

results-definition.xml

<?xml version="1.0"?>
<!--Phoronix Test Suite v10.8.4-->
<PhoronixTestSuite>
  <ResultsParser>
    <OutputTemplate>Throughput (items/sec): #_RESULT_#</OutputTemplate>
    <ResultScale>items/sec</ResultScale>
    <ResultProportion>HIB</ResultProportion>
  </ResultsParser>
  <ResultsParser>
    <OutputTemplate>Latency Mean (ms/batch): #_RESULT_#</OutputTemplate>
    <ResultScale>ms/batch</ResultScale>
    <ResultProportion>LIB</ResultProportion>
  </ResultsParser>
</PhoronixTestSuite>

test-definition.xml

<?xml version="1.0"?>
<!--Phoronix Test Suite v10.8.4-->
<PhoronixTestSuite>
  <TestInformation>
    <Title>Neural Magic DeepSparse</Title>
    <AppVersion>1.5</AppVersion>
    <Description>This is a benchmark of Neural Magic's DeepSparse using its built-in deepsparse.benchmark utility and various models from their SparseZoo (https://sparsezoo.neuralmagic.com/).</Description>
    <ResultScale>items/sec</ResultScale>
    <Proportion>HIB</Proportion>
    <TimesToRun>3</TimesToRun>
  </TestInformation>
  <TestProfile>
    <Version>1.5.0</Version>
    <SupportedPlatforms>Linux</SupportedPlatforms>
    <SoftwareType>Benchmark</SoftwareType>
    <TestType>System</TestType>
    <License>Free</License>
    <Status>Verified</Status>
    <ExternalDependencies>python</ExternalDependencies>
    <InstallRequiresInternet>TRUE</InstallRequiresInternet>
    <EnvironmentSize>9200</EnvironmentSize>
    <ProjectURL>https://neuralmagic.com/deepsparse-engine/</ProjectURL>
    <RepositoryURL>https://github.com/neuralmagic/deepsparse</RepositoryURL>
    <Maintainer>Michael Larabel</Maintainer>
    <SystemDependencies>pip3</SystemDependencies>
  </TestProfile>
  <TestSettings>
    <Default>
      <PostArguments> -t 30 -w 5</PostArguments>
    </Default>
    <Option>
      <DisplayName>Model</DisplayName>
      <Identifier>model</Identifier>
      <Menu>
        <Entry>
          <Name>NLP Text Classification, BERT base uncased SST2</Name>
          <Value>zoo:nlp/text_classification/bert-base/pytorch/huggingface/sst2/base-none</Value>
        </Entry>
        <Entry>
          <Name>NLP Text Classification, DistilBERT mnli</Name>
          <Value>zoo:nlp/text_classification/distilbert-none/pytorch/huggingface/mnli/base-none</Value>
        </Entry>
        <Entry>
          <Name>CV Classification, ResNet-50 ImageNet</Name>
          <Value>zoo:cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/base-none</Value>
        </Entry>
        <Entry>
          <Name>NLP Token Classification, BERT base uncased conll2003</Name>
          <Value>zoo:nlp/token_classification/bert-base/pytorch/huggingface/conll2003/base-none</Value>
        </Entry>
        <Entry>
          <Name>NLP Question Answering, BERT base uncased SQuaD 12layer Pruned90</Name>
          <Value>zoo:nlp/question_answering/bert-base/pytorch/huggingface/squad/12layer_pruned90-none</Value>
        </Entry>
        <Entry>
          <Name>CV Detection, YOLOv5s COCO</Name>
          <Value>zoo:cv/detection/yolov5-s/pytorch/ultralytics/coco/base-none</Value>
        </Entry>
        <Entry>
          <Name>NLP Document Classification, oBERT base uncased on IMDB</Name>
          <Value>zoo:nlp/document_classification/obert-base/pytorch/huggingface/imdb/base-none</Value>
        </Entry>
        <Entry>
          <Name>NLP Sentiment Analysis, 80% Pruned Quantized BERT Base Uncased</Name>
          <Value>zoo:nlp/sentiment_analysis/bert-base/pytorch/huggingface/sst2/12layer_pruned90-none</Value>
        </Entry>
        <Entry>
          <Name>CV Segmentation, 90% Pruned YOLACT Pruned</Name>
          <Value>zoo:cv/segmentation/yolact-darknet53/pytorch/dbolya/coco/pruned90-none</Value>
        </Entry>
      </Menu>
    </Option>
    <Option>
      <DisplayName>Scenario</DisplayName>
      <Identifier>scenario</Identifier>
      <ArgumentPrefix>--scenario </ArgumentPrefix>
      <Menu>
        <Entry>
          <Name>Synchronous Single-Stream</Name>
          <Value>sync</Value>
        </Entry>
        <Entry>
          <Name>Asynchronous Multi-Stream</Name>
          <Value>async</Value>
        </Entry>
      </Menu>
    </Option>
  </TestSettings>
</PhoronixTestSuite>