Core (2.6.0 ==> 2.7.0)
Features
- [core] MatMul broadcasting. (#4100).
- [webgl] Modularized mean kernel with custom shader. (#4033).
- Added mirrorPad kernel (#3951). Thanks, @AbdallaGomaa.
Bug fixes
- replace instanceof Promise with check of then function on the object (#4122).
- [webgl] Bump types dependency and remove from core. (#4109).
- Revert "[core] Make matMul accept mismatching batch dimensions when one input has batch 1." (#4099).
Development
Documentation
- Fix documentation description (#4097). Thanks, @GantMan.
- Fix docs - camelCase (#4089). Thanks, @GantMan.
Misc
- Update monorepo to 2.7.0. (#4137).
- Fix gradient for min and max (#4136).
- [core] Make matMul accept mismatching batch dimensions when one input has batch 1. (#3917).
- Modularize depthwise conv ops. (#4078).
- Modularize more conv ops. (#4075).
- Implement HashTable ops (#4060).
- e2e custom bundling for dense models. (#4042).
Data (2.6.0 ==> 2.7.0)
Bug fixes
- replace instanceof Promise with check of then function on the object (#4122).
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
Layers (2.6.0 ==> 2.7.0)
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
- Speed up builds. (#4072).
- Implement HashTable ops (#4060).
Converter (2.6.0 ==> 2.7.0)
Features
- [core] MatMul broadcasting. (#4100).
- Added mirrorPad kernel (#3951). Thanks, @AbdallaGomaa.
Bug fixes
- replace instanceof Promise with check of then function on the object (#4122).
- support partial shape for transform graph (#4057).
Documentation
- Properly escape wildcard asterisks (#4053). Thanks, @josephrocca.
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
- Implement HashTable ops (#4060).
- Modularize conv2D, fusedConv2D (#4059).
- Python converter bugfix (#4037). Thanks, @patlevin.
Node (2.6.0 ==> 2.7.0)
Features
- [core] MatMul broadcasting. (#4100).
- Add implementation for inputs and outputs getters to SavedModel (#4036). Thanks, @patlevin.
- Added mirrorPad kernel (#3951). Thanks, @AbdallaGomaa.
Development
- Modularize node backend kernels (#4119).
- Modularize Greater, GreaterEqual, Less, LessEqual, ... (#4106).
Documentation
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
- Speed up builds. (#4072).
Wasm (2.6.0 ==> 2.7.0)
Features
- [core] MatMul broadcasting. (#4100).
- [webgl] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3956).
Bug fixes
- [wasm] Fix bundler config and change module entry point in package.json. (#4121).
Documentation
- Fix broken all_kernels.ts URL (#4111). Thanks, @josephrocca.
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
- e2e custom bundling for dense models. (#4042).
Cpu (2.6.0 ==> 2.7.0)
Features
- [core] MatMul broadcasting. (#4100).
- Added mirrorPad kernel (#3951). Thanks, @AbdallaGomaa.
- [webgl] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3956).
Bug fixes
- Revert "[core] Make matMul accept mismatching batch dimensions when one input has batch 1." (#4099).
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
- Modularize Fill. (#4096).
- [core] Make matMul accept mismatching batch dimensions when one input has batch 1. (#3917).
- Reorder kernel orders. (#4082).
- Modularize depthwise conv ops. (#4078).
- Modularize more conv ops. (#4075).
- Modularize conv2D, fusedConv2D (#4059).
- Make generated files more linter friendly (#4063).
- e2e custom bundling for dense models. (#4042).
- Remove "kernelFunc" suffix from all kernel func names. Thanks, @jinjingforever.
Webgl (2.6.0 ==> 2.7.0)
Features
- [core] MatMul broadcasting. (#4100).
- [webgl] Modularized mean kernel with custom shader. (#4033).
- Added mirrorPad kernel (#3951). Thanks, @AbdallaGomaa.
- [webgl] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3956).
Bug fixes
- [webgl] Bump types dependency and remove from core. (#4109).
- Revert "[core] Make matMul accept mismatching batch dimensions when one input has batch 1." (#4099).
- [webgl] Fix uninitialized zeros tensors, fix dtype upcasting. (#4064).
Documentation
- add license header (#4046).
Misc
- Update release branch tfjs_2.7.0 lock files.
- Update monorepo to 2.7.0. (#4137).
- [core] Make matMul accept mismatching batch dimensions when one input has batch 1. (#3917).
- Speed up builds. (#4072).
- use webgl1 testEnv flags as registered (#4065).
- Remove "kernelFunc" suffix from all kernel func names. Thanks, @jinjingforever.