Models

Each model in the SparseZoo has a specific stub that identifies it. The stubs are made up of the following structure:

DOMAIN/SUB_DOMAIN/ARCHITECTURE{-SUB_ARCHITECTURE}/FRAMEWORK/REPO/DATASET{-TRAINING_SCHEME}/SPARSE_NAME-SPARSE_CATEGORY-{SPARSE_TARGET}

The properties within each model stub are defined as the following:

Model Property Definition Examples
DOMAIN The type of solution the model is architected and trained for cv, nlp
SUB_DOMAIN The sub type of solution the model is architected and trained for classification, segmentation
ARCHITECTURE The name of the guiding setup for the network's graph resnet_v1, mobilenet_v1
SUB_ARCHITECTURE (optional) The scaled version of the architecture such as width or depth 50, 101, 152
FRAMEWORK The machine learning framework the model was defined and trained in pytorch, tensorflow_v1
REPO The model repository the model and baseline weights originated from sparseml, torchvision
DATASET The dataset the model was trained on imagenet, cifar10
TRAINING_SCHEME (optional) A description on how the model was trained augmented, lower_lr
SPARSE_NAME An overview of what was done to sparsify the model base, pruned, quant (quantized), pruned_quant, arch (architecture modified)
SPARSE_CATEGORY Descriptor on the degree to which the model is sparsified as compared with the baseline metric none, conservative (100% baseline), moderate (>= 99% baseline), aggressive (< 99%)
SPARSE_TARGET (optional) Descriptor for the target environment the model was sparsified for disk, edge, deepsparse, gpu

The contents of each model are made up of the following:

  • model.md: The model card containing metadata, descriptions, and information for the model.

  • model.onnx: The ONNX representation of the model’s graph.

  • model.onnx.tar.gz: A compressed format for the ONNX file. Currently ONNX does not support sparse tensors and quantized sparse tensors well for compression.

  • [FRAMEWORK]/model.[EXTENSION]: The native ML framework file(s) for the model in which it was originally trained. Such as PyTorch, Keras, TensorFlow V1

  • recipes/original.[md|yaml]: The original sparsification recipe used to create the model.

  • recipes/[NAME].[md|yaml]: Additional sparsification recipes that can be used with the model such as transfer learning.

  • sample-originals: The original sample data without any preprocessing for use with the model.

  • sample-inputs: The sample data after pre processing for use with the model.

  • sample-outputs: The outputs after running the sample inputs through the model.

  • sample-labels: The labels that classify the sample inputs.

Image Classification

Model Tag Validation Baseline Metric
cv/classification/efficientnet-b0/pytorch/sparseml/imagenet/base-none 77.3% top1 accuracy
cv/classification/efficientnet-b0/pytorch/sparseml/imagenet/arch-moderate 76.5% top1 accuracy
cv/classification/efficientnet-b4/pytorch/sparseml/imagenet/base-none 83.0% top1 accuracy
cv/classification/efficientnet-b4/pytorch/sparseml/imagenet/arch-moderate 82.1% top1 accuracy
cv/classification/inception_v3/pytorch/sparseml/imagenet/base-none 77.4% top1 accuracy
cv/classification/inception_v3/pytorch/sparseml/imagenet/pruned-conservative 77.4% top1 accuracy
cv/classification/inception_v3/pytorch/sparseml/imagenet/pruned-moderate 76.6% top1 accuracy
cv/classification/mnistnet/pytorch/sparseml/mnist/base-none 99.4% top1 accuracy
cv/classification/mobilenet_v1-1.0/pytorch/sparseml/imagenet/base-none 70.9% top1 accuracy
cv/classification/mobilenet_v1-1.0/pytorch/sparseml/imagenet/pruned-conservative 70.9% top1 accuracy
cv/classification/mobilenet_v1-1.0/pytorch/sparseml/imagenet/pruned-moderate 70.1% top1 accuracy
cv/classification/mobilenet_v1-1.0/pytorch/sparseml/imagenet/pruned_quant-moderate 70.1% top1 accuracy
cv/classification/mobilenet_v2-1.0/pytorch/sparseml/imagenet/base-none 71.9% top1 accuracy
cv/classification/resnet_v1-101/keras/sparseml/imagenet/base-none 77.4% top1 accuracy
cv/classification/resnet_v1-101/keras/sparseml/imagenet/pruned-moderate 76.6% top1 accuracy
cv/classification/resnet_v1-101/pytorch/sparseml/imagenet/base-none 77.4% top1 accuracy
cv/classification/resnet_v1-101/pytorch/sparseml/imagenet/pruned-moderate 76.6% top1 accuracy
cv/classification/resnet_v1-101/pytorch/torchvision/imagenet/base-none 76.6% top1 accuracy
cv/classification/resnet_v1-101_2x/pytorch/sparseml/imagenet/base-none 78.8% top1 accuracy
cv/classification/resnet_v1-101_2x/pytorch/torchvision/imagenet/base-none 78.8% top1 accuracy
cv/classification/resnet_v1-152/keras/sparseml/imagenet/base-none 78.3% top1 accuracy
cv/classification/resnet_v1-152/keras/sparseml/imagenet/pruned-moderate 77.5% top1 accuracy
cv/classification/resnet_v1-152/pytorch/sparseml/imagenet/base-none 78.3% top1 accuracy
cv/classification/resnet_v1-152/pytorch/sparseml/imagenet/pruned-moderate 77.5% top1 accuracy
cv/classification/resnet_v1-152/pytorch/torchvision/imagenet/base-none 77.5% top1 accuracy
cv/classification/resnet_v1-18/pytorch/sparseml/imagenet/base-none 69.8% top1 accuracy
cv/classification/resnet_v1-18/pytorch/sparseml/imagenet/pruned-conservative 69.8% top1 accuracy
cv/classification/resnet_v1-18/pytorch/torchvision/imagenet/base-none 69.8% top1 accuracy
cv/classification/resnet_v1-20/keras/sparseml/cifar_10/base-none 91.3% top1 accuracy
cv/classification/resnet_v1-34/pytorch/sparseml/imagenet/base-none 73.3% top1 accuracy
cv/classification/resnet_v1-34/pytorch/sparseml/imagenet/pruned-conservative 73.3% top1 accuracy
cv/classification/resnet_v1-34/pytorch/torchvision/imagenet/base-none 73.3% top1 accuracy
cv/classification/resnet_v1-50/keras/sparseml/imagenet/base-none 76.1% top1 accuracy
cv/classification/resnet_v1-50/keras/sparseml/imagenet/pruned-conservative 76.1% top1 accuracy
cv/classification/resnet_v1-50/keras/sparseml/imagenet/pruned-moderate 75.3% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/base-none 76.1% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/pruned-conservative 76.1% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/pruned-moderate 75.3% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenet/pruned_quant-moderate 75.4% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenet-augmented/pruned_quant-aggressive 76.1% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenette/base-none 99.9% top1 accuracy
cv/classification/resnet_v1-50/pytorch/sparseml/imagenette/pruned-conservative 99.9% top1 accuracy
cv/classification/resnet_v1-50/pytorch/torchvision/imagenet/base-none 99.9% top1 accuracy
cv/classification/resnet_v1-50/pytorch/torchvision/imagenette/pruned-conservative 99.9% top1 accuracy
cv/classification/resnet_v1-50_2x/pytorch/sparseml/imagenet/base-none 78.1% top1 accuracy
cv/classification/resnet_v1-50_2x/pytorch/torchvision/imagenet/base-none 78.1% top1 accuracy
cv/classification/vgg-11/pytorch/sparseml/imagenet/base-none 69.0% top1 accuracy
cv/classification/vgg-11/pytorch/sparseml/imagenet/pruned-moderate 68.3% top1 accuracy
cv/classification/vgg-11/pytorch/torchvision/imagenet/base-none 68.3% top1 accuracy
cv/classification/vgg-11_bn/pytorch/sparseml/imagenet/base-none 70.4% top1 accuracy
cv/classification/vgg-11_bn/pytorch/torchvision/imagenet/base-none 70.4% top1 accuracy
cv/classification/vgg-13/pytorch/sparseml/imagenet/base-none 69.9% top1 accuracy
cv/classification/vgg-13/pytorch/torchvision/imagenet/base-none 69.9% top1 accuracy
cv/classification/vgg-13_bn/pytorch/sparseml/imagenet/base-none 71.5% top1 accuracy
cv/classification/vgg-13_bn/pytorch/torchvision/imagenet/base-none 71.5% top1 accuracy
cv/classification/vgg-16/pytorch/sparseml/imagenet/base-none 71.6% top1 accuracy
cv/classification/vgg-16/pytorch/sparseml/imagenet/pruned-conservative 71.6% top1 accuracy
cv/classification/vgg-16/pytorch/sparseml/imagenet/pruned-moderate 70.8% top1 accuracy
cv/classification/vgg-16/pytorch/torchvision/imagenet/base-none 70.8% top1 accuracy
cv/classification/vgg-16_bn/pytorch/sparseml/imagenet/base-none 71.6% top1 accuracy
cv/classification/vgg-16_bn/pytorch/torchvision/imagenet/base-none 71.6% top1 accuracy
cv/classification/vgg-19/pytorch/sparseml/imagenet/base-none 72.4% top1 accuracy
cv/classification/vgg-19/pytorch/sparseml/imagenet/pruned-moderate 71.7% top1 accuracy
cv/classification/vgg-19/pytorch/torchvision/imagenet/base-none 71.7% top1 accuracy
cv/classification/vgg-19_bn/pytorch/sparseml/imagenet/base-none 74.2% top1 accuracy
cv/classification/vgg-19_bn/pytorch/torchvision/imagenet/base-none 74.2% top1 accuracy

Object Detection

Model Tag Validation Baseline Metric
cv/detection/ssd-resnet50_300/pytorch/sparseml/coco/base-none 42.7 [email protected]
cv/detection/ssd-resnet50_300/pytorch/sparseml/coco/pruned-moderate 41.8 [email protected]
cv/detection/ssd-resnet50_300/pytorch/sparseml/voc/base-none 52.2 [email protected]
cv/detection/ssd-resnet50_300/pytorch/sparseml/voc/pruned-moderate 51.5 [email protected]
cv/detection/yolo_v3-spp/pytorch/ultralytics/coco/base-none 64.2 [email protected]
cv/detection/yolo_v3-spp/pytorch/ultralytics/coco/pruned-aggressive_97 62.4 [email protected]
cv/detection/yolo_v3-spp/pytorch/ultralytics/coco/pruned_quant-aggressive_94 60.5 [email protected]