SparseZoo v0.7.0

New Features:

• Pruned-Quantized BERT models on SQuAD added.
• Support added for
  • Python 3.9.
  • ONNX versions 1.8 - 1.10.

Changes:

• None

Resolved Issues:

• 403 errors no longer occasionally occur when trying to download models.

Known Issues:

• None

SparseZoo v0.6.0

New Features:

• Support added for batched iteration of a dataset.

Changes:

• None

Resolved Issues:

• Load_numpy_list API no longer crashes when loading from a directory.

Known Issues:

• None

SparseZoo v0.5.1 Patch Release

This is a patch release for 0.5.0 that contains the following changes:

• version updated for consistency to reflect DeepSparse repo hotfix.

SparseZoo v0.5.0

New Features:

• Alpha website sparsezoo.neuralmagic.com instrumented and deployed for visual interaction with the repository.

Changes:

• None

Resolved Issues:

• API requests for models on Windows systems now working consistently.

Known Issues:

• None

SparseZoo v0.4.0

New Features:

• None

Changes:

• Readme redesigned for better clarity on the repository's purpose.

Resolved Issues:

• Neural Magic dependencies upgrade to intended bug versions instead of minor versions.

Known Issues:

• None

SparseZoo v0.3.1 Patch Release

This is a patch release for 0.3.0 that contains the following changes:

• Docs updated for new Discourse and Slack links
• Table UI removed from the codebase and package to make the build process more consistent and size smaller
  -- Table UI is now accessible at https://sparsezoo.neuralmagic.com/tables/models/cv/detection

SparseZoo v0.3.0

New Features:

• SparseZoo table UI added for finding available recipes and models in the zoo.
• APIs added for looking up similar recipes and getting the base model associated with a recipe.
• Convenience methods added to the Model and Recipe class to lookup related objects to the current instance.

Changes:

• Version changed to be loaded from version.py file, default build on branches is now nightly.
• Removed all references to optim and optimization in favor of sparse and sparsification.

Resolved Issues:

• None

Known Issues:

• None

SparseZoo v0.2.0

New Features:

• YOLOv3 base, pruned, pruned-quantized models added for ultralytics/yolov5 repo.
• Keras pruned resnet models implemented.
• Ability to load recipes from SparseZoo stubs available.
• Ability to load the base weights for a sparsified model available.

Changes:

• Default log level changed to warn.
• SparseZoo backend improved for faster deployments and more uptime.

Resolved Issues:

• Docs and readme corrections made for minor issues and broken links.
• Makefile no longer deletes files for docs compilation and cleaning.

Known Issues:

• None

SparseZoo v0.1.1 Patch Release

This is a patch release for 0.1.0 that contains the following changes:

• Docs updates: tagline, overview, update to use sparsification for verbiage
• ResNet-50 pruned_quant moderate model added for PyTorch SparseML
• Functionality added for loading recipes and corresponding model weights from zoo stubs
• Nightly build dependencies now match on major.minor and not full version
• Change default log level to warn from info

SparseZoo v0.1.0 First GitHub Release

Welcome to our initial release on GitHub! Older release notes can be found here.

New Features:

• Python API is available to interact with models in the SparseZoo.
• CLI added to interact with models in the SparseZoo.
• model_download Jupyter Notebook added.
• Test cases introduced for published models.
• MakeFile flows and utilities implemented for GitHub repo structure.

Changes:

• Software packaging updated to reflect new GitHub distribution channel, from file naming conventions to license enforcement removal.

Resolved Issues:

• None

Known Issues:

• None