Releases: facebookresearch/playtorch
0.2.4
0.2.4 contains the following changes
Learn more about the PlayTorch name and the PlayTorch app in our announcement blog post: https://pytorch.org/blog/introducing-the-playtorch-app/
react-native-pytorch-core
- Tensor Indexing API for set tensor (68d976f by @raedle)
- Fix issue with JNI Env not available (0723cc0 by @raedle)
- Fix issue with JNI Env not available for audio (c0b2026 by @raedle)
Full Changelog: v0.2.3...v0.2.4
0.2.3
0.2.3 contains the following notable changes plus many smaller improvements and fixes.
Learn more about the PlayTorch name and the PlayTorch app in our announcement blog post: https://pytorch.org/blog/introducing-the-playtorch-app/
💫 The 0.2.0 series introduced new JavaScript interfaces to PyTorch APIs for flexible data processing and inference that replace MobileModel.execute
and Live Spec JSON. See the core package's README for example usage.
react-native-pytorch-core
- Add tensor.matmul to PyTorch SDK for JSI (7f633ca by @zrfisher)
- Update model urls from previous GitHub repo
pytorch/live
tofacebookresearch/playtorch
(edf43de by @raedle) - Fix minor compiler warnings (bd9d13d by @raedle)
- Add pragma marks to ignore
deprecated-delcarations
surfaced by C++17 compiler. This was needed because PyTorch Mobile is C++14 (20842f6 by @raedle) - Introduce new image toBlob API function signature (#130) (ffa4ed6 by @raedle)
torch.jit._load_for_mobile
with device and extra files options (#141) (e249116 by @raedle)- Patch failing issue with RN 0.64.3 (#157) (5e6b2c0 by @raedle)
- Enforce 3 argument requirement for randint (02c37a3 by @neildhar)
- upgrade prettier to 2.7.1 (74a3ed8 by @bradzacher)
Breaking Changes
react-native-template-pytorch-live
-
Update model urls from previous GitHub repo
pytorch/live
tofacebookresearch/playtorch
(bfd6a97 by @raedle) -
Note:
react-native-template-pytorch-live
is deprecated. Instead, follow React Native’s Environment Setup guide. To usereact-native-pytorch-core
in an existing app follow the documentation Add PlayTorch to Existing App or in an Expo managed app usenpx expo install react-native-pytorch-core
.
Thank you to our contributors @zrfisher, @neildhar, and @bradzacher.
Full Changelog: v0.2.2...v0.2.3
0.2.2
0.2.2 contains the following improvements and removes some deprecated features.
💫 The 0.2.0 series introduced new JavaScript interfaces to PyTorch APIs for flexible data processing and inference that replace MobileModel.execute
and Live Spec JSON. See the core package's README for example usage.
react-native-pytorch-core
- Remove “Live Spec” APIs. Instead, use the JavaScript interfaces to PyTorch APIs introduced with 0.2.0 (#111)
- Force same orientation on ImageUtil.fromFile (#124)
- Extend Tensor.topk() function parameters to match PyTorch API
- Fix a bug with Tensor indexing APIs (#118)
react-native-template-pytorch-live
- Remove related “slim template” package
react-native-template-ptl
. Instead, get started with the standard React Native CLI, Expo, or Expo EAS
torchlive-cli
- torchlive-cli was deprecated in 0.2.1 and has been removed in this release. Instead, follow the instructions in the Manual Environment Setup guide (#121)
Thank you to our contributors @cjfghk5697, @liuyinglao, @raedle, @reedless, and Zeinab Sadeghipour Kermani
Full Changelog: v0.2.1...v0.2.2
0.2.1
0.2.1 contains the following notable changes plus many smaller improvements and fixes.
This is the first release under the new name "PlayTorch" which replaces "PyTorch Live". The repository now lives at https://github.com/facebookresearch/playtorch
Learn more about the PlayTorch name and the PlayTorch app in our announcement blog post: https://pytorch.org/blog/introducing-the-playtorch-app/
💫 The 0.2.0 series introduced new JavaScript interfaces to PyTorch APIs for flexible data processing and inference that replace MobileModel.execute
and Live Spec JSON. See the core package's README for example usage.
react-native-pytorch-core
- Support passing general JavaScript types (e.g., Array, String, Object) to
Module.forward
, and automatically unpack result types (IValue
in C++) to JavaScript types - Support for React Native 0.68, 0.69
- Fix Expo Config Plugin: now the apps generated by
expo prebuild
are ready to compile - Fix iOS
<Canvas>
element to match the (correct) Android behavior when callingCanvasRenderingContext2D.invalidate()
to repaint - Add new PyTorch JavaScript API wrappers:
torch.cat
,torch.full
,torch.linspace
,torch.logspace
,torch.randperm
,torch.randn
,Tensor.argmin
,Tensor.expand
, andTensor.flip
- Add support for RGBA and grayscale to
media.imageFromTensor
- Deprecate
media.imageFromBlob
in favor ofmedia.imageFromTensor
- Deprecate
MobileModel.execute
,MobileModel.preload
,MobileModel.unload
, and Live Spec in general. Migrate away from Live Spec to usetorch.jit._loadForMobile
- Fix toDictStringKey (#99) in legacy Live Spec
react-native-template-pytorch-live
- Update documentation references from PyTorch Live to PlayTorch
torchlive-cli
- Note: torchlive-cli is deprecated. Instead, follow React Native’s Environment Setup guide. To use the default
torchlive-cli
template, pass the--template
flag during setup:npx react-native init AwesomeProject --template react-native-template-pytorch-live
Thank you to our contributors @ansonsyfang, @bhadreshpsavani, @chrisklaiber, @cjfghk5697, @clarksandholtz, @justinhaaheim, @liuyinglao, @michaelkulinich, @raedle, @simpleton, @ta211, Kyle Into, Lucca Bertoncini, Prakhar Sahay, Shushanth Madhubalan, Vladimir Pinchuk, and Zonggen Yi.
Full Changelog: v0.2.0...v0.2.1
0.2.0
0.2.0 is a major update with various bug fixes and the following notable changes. For an introduction to the new platform's capabilities, see our PlayTorch announcement blog post and new home at https://playtorch.dev
react-native-pytorch-core
- 💫 New JavaScript interfaces to PyTorch APIs for flexible data processing and inference. This replaces
MobileModel.execute
and Live Spec JSON. See the core package's README with example usage - Example code now uses the JavaScript interfaces to PyTorch APIs
- New image APIs to convert from image to blob to tensor and back
- Expand model support to include, for example, YOLOv5, DeepLabV3, Fast Neural Style, Wav2Vec2
- Update PyTorch Mobile dependency from 1.10 to 1.12
- Improve support for React Native 0.66+
- Improve support for Apple Silicon
react-native-template-pytorch-live
- Remove dependency on Python and locally created models via
make_models.py
. Instead, the pre-exported models are loaded from the network - Example code now uses the JavaScript interfaces to PyTorch APIs
torchlive-cli
- Note: torchlive-cli will be deprecated in an upcoming release. Instead, follow React Native’s Environment Setup guide. To use the default
torchlive-cli
template, pass the--template
flag during setup:npx react-native init AwesomeProject --template react-native-template-pytorch-live
Thank you to our contributors @ansonsyfang, @ap190, @chrisklaiber, @clarksandholtz, @justinhaaheim, @krazykalman, @liuyinglao, @metawaku, @pavlos-chatzisavvas, @pd21989, @raedle, @ta211, @williamngan, Alp Genc, Bode Sule, Isaac Mosebrook, Minji Kim, and Thiago Roscia
Full Changelog: v0.1.3...v0.2.0
0.2.0-rc.3
Full Changelog: v0.2.0-rc.2...v0.2.0-rc.3
0.2.0-rc.2
Full Changelog: v0.2.0-rc.1...v0.2.0-rc.2
0.2.0-rc.1
Full Changelog: v0.2.0-rc.0...v0.2.0-rc.1
0.2.0-rc.0
Full Changelog: https://github.com/pytorch/live/commits/v0.2.0-rc.0
v0.1.3
0.1.3 is out with changes:
Fixed
- Remove unused experimental annotation (bee9aec) @raedle
- Fix the detox test in the template app (239070f) @liuyinglao
- Hardcoding length constraint into bert_qa model spec (804f413) @clarksandholtz
- Create an AVD for the M1 Macs (avoids emulator failure) (#24) (3c096ea) @aaronespasa
- Some cleanups to the model spec documentation (#27 ) (620b4d5) @mdwelsh
- NPM publish to @latest only when release type is released (f9a399d) @liuyinglao
- Fix encoding issue on Windows 11 (698132e) @raedle
Features
- Add fromJSRef method to ImageUtil (#31) (4825d28) @Sxela
- Add e2e tests for JSI Configuration (9f5318a) @liuyinglao
- Created slim template (#28) (a52ab6d) @clarksandholtz
- Setup Pytorch Live to run cpp function with JSI (80e7bef) @liuyinglao
- Run make_models.py in post-init script on Windows (b580458) @raedle
Full Changelog: v0.1.2...v0.1.3