Core (3.16.0 ==> 3.17.0)
Features
- Add LowerBound and UpperBound kernels (#6356).
Bug fixes
- Fix CPU fusedConv2d (bias add) for NCHW format (#6379). Thanks, @Linchenn.
- Fix WebGL Conv2d for NCHW format (#6340). Thanks, @Linchenn.
- : #6284 [webgpu: support NCHW dataFormat for conv2d] (#6293). Thanks, @qjia7.
Misc
- Update monorepo to 3.17.0. (#6405).
- Add string support for SparseToDense (#6397).
- Improve link_package speed (#6376).
- Add euclideanNorm op. (#6366).
- Fix tfjs-core .d.ts files missing webgpu refrence types (#6386).
- webgpu: support dataToGPU api (#6329). Thanks, @xhcao.
- Fix WebGL-conv2dByMatMul for NCHW with multiple batches (#6369). Thanks, @Linchenn.
- Change makeTensorFromTensorId to makeTensorFromTensorInfo (#6355). Thanks, @qjia7.
- Add Kernel SearchSorted for CPU and WebGL backend (#6349).
Data (3.16.0 ==> 3.17.0)
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
Layers (3.16.0 ==> 3.17.0)
Bug fixes
- check for null before incorporating scale and offset (#6383).
- Serialize UpSampling2D 'interpolation' option (#6360). Thanks, @debagger.
Performance
- Co-authored-by: Ping Yu 4018+pyu10055@users.noreply.github.com [sort feedDict] (#6327). Thanks, @Linchenn.
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
- Improve link_package speed (#6376).
Converter (3.16.0 ==> 3.17.0)
Features
- Add tensorListConcatV2 support. (#6351).
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
- Upgrade rules_nodejs to 5.4.2 (#6404).
- Improve link_package speed (#6376).
- Add euclideanNorm op. (#6366).
- Add support for LowerBound and UpperBound conversion (#6364).
Node (3.16.0 ==> 3.17.0)
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
- Improve link_package speed (#6376).
- [node] Fix memory leak from string tensors (#6371).
- [node] Use makeTensorFromTensorInfo instead of makeTensorFromId (#6361).
- Add Kernel SearchSorted for CPU and WebGL backend (#6349).
Wasm (3.16.0 ==> 3.17.0)
Security
- Bump async from 2.6.3 to 2.6.4 in /tfjs-backend-wasm/starter/webpack (#6343). Thanks, @dependabot[bot].
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
- Improve link_package speed (#6376).
Cpu (3.16.0 ==> 3.17.0)
Bug fixes
Documentation
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
- Add string support for SparseToDense (#6397).
- Fix bufferSync return type (#6362).
- Improve link_package speed (#6376).
- Change makeTensorFromTensorId to makeTensorFromTensorInfo (#6355). Thanks, @qjia7.
- Add Kernel SearchSorted for CPU and WebGL backend (#6349).
Webgl (3.16.0 ==> 3.17.0)
Bug fixes
- Fix CPU fusedConv2d (bias add) for NCHW format (#6379). Thanks, @Linchenn.
- Fix WebGL Conv2d (PReLU computation) for NCHW format (#6365). Thanks, @Linchenn.
- Fix WebGL Conv2d for NCHW format (#6340). Thanks, @Linchenn.
Misc
- Update lockfiles branch tfjs_3.17.0_lockfiles lock files. (#6415).
- Update monorepo to 3.17.0. (#6405).
- Add string support for SparseToDense (#6397).
- Fix bufferSync return type (#6362).
- Fix WebGL-conv2dByMatMul for NCHW with multiple batches (#6369). Thanks, @Linchenn.
- Decrease values in WebGL tests to fit into WebGL1's range (#6372). Thanks, @Linchenn.
- Change makeTensorFromTensorId to makeTensorFromTensorInfo (#6355). Thanks, @qjia7.
- Add Kernel SearchSorted for CPU and WebGL backend (#6349).
- [webgl] merge CumProdProgram and CumSumProgram (#6325). Thanks, @axinging.