Releases: PerfectlySoft/Perfect-TensorFlow
Releases · PerfectlySoft/Perfect-TensorFlow
Fixing Namespace conflicts between PerfectLib & Apple's Swift Protocol Buffer Compiler
Pre-release
1.3.1 Fixing executable namespace conflicts issue between Perfect and Apple…
Upgrading to TensorFlow 1.3.0.
Features:
- Fixing QInt16 size issue
- Gradient has new namespaces
- Adding Devices to Session for distributive computation.
Upgrading to TensorFlow v1.2.1 binary release
graph.addGradients()
is active now.
Adding Matrix Tensor with auto shape / auto flat functionality
1.2.1 Adding Matrix Tensor with auto shape / auto flat functionality.
Fully Upgrading to TensorFlow 1.2.0
- Based on TensorFlow 1.2.0 binary distribution.
- Fixing memory leak between sessions.
- Adding Gradients() to graph
- Tensor is now comparable.
- Swift 3.1 / 4.0 Compatible.
- Adding Demo of Computer Vision
- Full English / Chinese Document Support.
Correcting Typos
1.1.14 Correcting typos.
Fixing tensor memory leaks after session.
According to https://github.com/tensorflow/tensorflow/issues/10997 Before this fix, tensors created by session.run() came to wild because of calling tensor.init(handler:) with tensor.autodestroy = false. The solution is patching and setting it to true value after creation.
Adding GUIDE; Adding loadModel() to Session.Runner()
1.1.12 Cleaning up GUIDE.md
Swift 3.1/4.0 Support; Fixing constant(index::) default zero.
Fixing constant(index::) automatically set to zero Also adding memory leak tests.
Adding compatibility of both Swift 3.1 / 4.0
There are still a few warnings in Swift 4.0 compilation, one from XCTest and one from Swift Protobuf.