TensorFlow.js 2.0.0

TensorFlow.js 2.0.0 is here! It is available on NPM as well as CDNs that sync from NPM.

It contains a few breaking changes, details of which are listed below, but at a high level these are some of the major changes in this release:

• The CPU and WebGL backends have moved out of tfjs-core into their own packages. tfjs-backend-cpu and tfjs-backend-webgl.
  • If you are not using the union package (i.e. you use tfjs-core, tfjs-layers, tfjs-converter, tfjs-data directly) you need to import these backends into your application to get the same behavior as before, or you can only import a single backend if that is all you need.
  • If you use the union package (i.e. tfjs), then nothing will need to change in your code as these backends are already included.
  • In some situations (e.g. running certain ops with small tensors) the WebGL backend can optionally fall back to the CPU backend, to keep this behavior, for those not using the union package you should include both the cpu and webgl backend in your application. We are working towards saving you this step in 3.x. If you primarily rely on the WebGL backend, you should test the performance with and without the CPU backend present and make the appropriate decision of what to include.
• In addition to our ES5 bundles, we are shipping new pre-compiled bundles with ES modules and ES2017 code:
  • tf.min.js & tf.es2017.min.js
  • tf-core.min.js & tf-core.es2017.min.js
  • tf-backend-cpu.min.js & tf-backend-cpu.es2017.min.js
  • tf-backend-webgl.min.js & tf-backend-webgl.es2017.min.js
  • tf-layers.min.js & tf-layers.es2017.min.js
  • tf-converter.min.js & tf-converter.es2017.min.js
  • tf-data.min.js & tf-data.es2017.min.js
• Our 'module' entry in package.json points to files with ES modules and ES2017 code. These are individual files corresponding to the source code of our libraries. This lays down a path to enable tree shaking and smaller bundles in apps.
  • If you have previously been using the files in the 'dist' subfolder of our NPM package, you may need to adjust your build system to consume them.
• Functionality that was deprecated in 1.x has been removed.
• In 2.x we are deprecating all the '*Strict' variants of ops. e.g. divStrict and mulStrict.

Core (1.7.4 ==> 2.0.0)

Features

• [wasm] Add split, sqrt kernels. (#3183).

Breaking changes

• Remove deprecated batchnormalization op (#3238).
• Update packages to compile to ESM modules by default (#3112).
• move webgl backend out of core (#3056).
• Move cpu backend out of tfjs-core (#3008).

Bug fixes

• [[core] Restrict size of conv test output to temporarily fix precision issues in Safari.] (#3187).
• [[core] Change inputs to matMul test to avoid precision issues in Safari as stopgap.] (#3142).
• fix nightly error on isMobile (#3139).
• [webgpu] Fix dilation and add stride 2 filter 3x3 same conv2d… (#2846). Thanks, @axinging.
• ensure isMobile() does not fail when navigator is not set. [ensure isMobile() does not fail when navigator is not set] (#3075).
• [core] Fix toPixels memLeak for TensorLike inputs. (#3073).
• Remove duplicate declaration. ([#3030](https://github.com/
• use the loadOptions to find the IOHandler (#3286)./pull/3030)).
• [[core] Restrict size of conv test output to temporarily fix precision issues in Safari.] (#3187).
• [[core] Change inputs to matMul test to avoid precision issues in Safari as stopgap.] (#3142).
• [core] Fix toPixels memLeak for TensorLike inputs. (#3073).

Development

• deprecate strict variants of ops (#3239).
• Add license to pad3d.ts. [Add license.] (#3027).
• Modularize max. (#2955).
• Modularize lrn. [Modularize lrn.] (#3218).
• Modularize sub. (#3021).
• Modularize matmul. [Modularize matmul.] (#3223).
• Modularize diag. [Modularize diag.] (#3217).
• Modularize conv ops. [[core] Modularize conv ops.] (#3120).
• Nightly test fixes [Nightly test fixes] (#3208).
• Modularize greaterEqual. [Modularize greaterEqual.] (#3192).
• Modularize lessEqual. [Modularize lessEqual.] (#3190).
• Modularize equal. [Modularize equal.] (#3189).
• Modularize less. [Modularize less.] (#3188).
• Fix browserstack iPhone browser issue. [Fix browserstack iPhone browser issue.] (#3204).
• Modularize greater. [Modularize greater.] (#3191).
• Modularize concat and split. [[core] Modularize concat and split.] (#3093).
• Remove use of chained ops. [Remove use of chained ops.] (#3058).
• Modularize notEqual. [[core]Modularize notEqual.] (#3091).
• update yarn.lock (#3090).
• fix nightly test for isMobile. [fix nightly test for isMobile] (#3079).
• Remove backward func from SquaredDifference. [Clean up. Remove backward func from SquaredDifference.] (#3055).
• Add this.throwIfDisposed() to modularized ops. [Add this.throwIfDisposed();] (#3044).
• Modularize depthToSpace. (#3277).
• Modularize spaceToBatchND. (#3273).
• Modularize batchToSpaceND. (#3272).

Misc

Data (1.7.4 ==> 2.0.0)

Breaking changes

• remove deprecated functionality (#3233).
• Update packages to compile to ESM modules by default (#3112).

Layers (1.7.4 ==> 2.0.0)

Breaking changes

• Remove deprecated batchnormalization op (#3238).
• Update packages to compile to ESM modules by default (#3112).
• move webgl backend out of core (#3056).
• Move cpu backend out of tfjs-core (#3008).

Converter (1.7.4 ==> 2.0.0)

Features

• added support select v2 op (#3154).

Breaking changes

• Update packages to compile to ESM modules by default (#3112).
• Move cpu backend out of tfjs-core (#3008).

Bug fixes

• fix tensor array gather not be able to handle indices longer than available tensors (#3157).
• remove unnecessary eager call (#3145).

Misc

• Switching to tf.io. for file access and loading savedmodels in eager model. (#3126). Thanks, @davidzats-eng.

Node (1.7.4 ==> 2.0.0)

Features

• Add support for napi version 6. [Add support for napi version 6] (#3242).

Breaking changes

• Remove deprecated batchnormalization op (#3238).
• [node] Remove deprecated createTypeOpAttr (#3235).
• Update packages to compile to ESM modules by default (#3112).

Bug fixes

• Fix node and nodegpu build. (#3248).

Development

• Modularize conv ops. [[core] Modularize conv ops.] (#3120).

Documentation

• correct typo on tf.node.tensorBoard doc (#3243). Thanks, @avindra.
• fix import in README.md (#3201). Thanks, @qwertyforce.

Misc

• Remove layers symlink in node-gpu. (#3031).

Wasm (1.7.4 ==> 2.0.0)

Features

• [wasm] Add split, sqrt kernels. (#3183).
• use instantiateWasm override and platform fetch (#3100).

Breaking changes

• Update packages to compile to ESM modu...

tfjs-2.0.0-rc.4 public release candidate

This is a public release candidate for TensorFlow.js 2.0. It is available on NPM as well as CDNs that sync from NPM. Since it is a release candidate you will need to opt into it by explicitly putting tfjs-2.0.0-rc.4 into your package.json or script tags.

It contains a few breaking changes, details of which are listed below, but at a high level these are some of the major changes in this release:

• The CPU and WebGL backends have moved out of tfjs-core into their own packages. tfjs-backend-cpu and tfjs-backend-webgl.
  • If you are not using the union package (i.e. you use tfjs-core, tfjs-layers, tfjs-converter, tfjs-data directly) you need to import these backends into your application to get the same behavior as before, or you can only import a single backend if that is all you need.
  • If you use the union package (i.e. tfjs), then nothing will need to change in your code as these backends are already included.
• In addition to our ES5 bundles, we are shipping new pre-compiled bundles with ES modules and ES2017 code:
  • tf.min.js & tf.es2017.min.js
  • tf-core.min.js & tf-core.es2017.min.js
  • tf-backend-cpu.min.js & tf-backend-cpu.es2017.min.js
  • tf-backend-webgl.min.js & tf-backend-webgl.es2017.min.js
  • tf-layers.min.js & tf-layers.es2017.min.js
  • tf-converter.min.js & tf-converter.es2017.min.js
  • tf-data.min.js & tf-data.es2017.min.js
• Our 'module' entry in package.json points to files with ES modules and ES2017 code. These are individual files corresponding to the source code of our libraries. This lays down a path to enable tree shaking and smaller bundles in apps.
  • If you have previously been using the files in the 'dist' subfolder of our NPM package, you may need to adjust your build system to consume them.
• Functionality that was deprecated in 1.x has been removed.
• In 2.x we are deprecating all the '*Strict' variants of ops. e.g. divStrict and mulStrict.

Core (1.7.4 ==> 2.0.0-rc.4)

Features

• [wasm] Add split, sqrt kernels. (#3183).

Breaking changes

• Remove deprecated batchnormalization op (#3238).
• Update packages to compile to ESM modules by default (#3112).
• move webgl backend out of core (#3056).
• Move cpu backend out of tfjs-core (#3008).

Bug fixes

• Restrict size of conv test output to temporarily fix precision issues in Safari. [[core] Restrict size of conv test output to temporarily fix precision issues in Safari.] (#3187).
• Change inputs to matMul test to avoid precision issues in Safari as stopgap. [[core] Change inputs to matMul test to avoid precision issues in Safari as stopgap.] (#3142).
• fix nightly error on isMobile (#3139).
• [webgpu] Fix dilation and add stride 2 filter 3x3 same conv2d… (#2846). Thanks, @axinging.
• ensure isMobile() does not fail when navigator is not set. [ensure isMobile() does not fail when navigator is not set] (#3075).
• [core] Fix toPixels memLeak for TensorLike inputs. (#3073).
• Remove duplicate declaration. [Remove duplicate declaration.] (#3030).

Development

• Add license to pad3d.ts. [Add license.] (#3027).
• Modularize max. (#2955).
• Modularize lrn. [Modularize lrn.] (#3218).
• Modularize sub. (#3021).
• Modularize matmul. [Modularize matmul.] (#3223).
• Modularize diag. [Modularize diag.] (#3217).
• Modularize conv ops. [[core] Modularize conv ops.] (#3120).
• Nightly test fixes [Nightly test fixes] (#3208).
• Modularize greaterEqual. [Modularize greaterEqual.] (#3192).
• Modularize lessEqual. [Modularize lessEqual.] (#3190).
• Modularize equal. [Modularize equal.] (#3189).
• Modularize less. [Modularize less.] (#3188).
• Fix browserstack iPhone browser issue. [Fix browserstack iPhone browser issue.] (#3204).
• Modularize greater. [Modularize greater.] (#3191).
• Modularize concat and split. [[core] Modularize concat and split.] (#3093).
• Remove use of chained ops. [Remove use of chained ops.] (#3058).
• Modularize notEqual. [[core]Modularize notEqual.] (#3091).
• update yarn.lock (#3090).
• fix nightly test for isMobile. [fix nightly test for isMobile] (#3079).
• Remove backward func from SquaredDifference. [Clean up. Remove backward func from SquaredDifference.] (#3055).
• Add this.throwIfDisposed() to modularized ops. [Add this.throwIfDisposed();] (#3044).

Misc

• deprecate strict variants of ops (#3239).

Data (1.7.4 ==> 2.0.0-rc.4)

Features

• [core] Modularize max. (#2955).

Breaking changes

• remove deprecation (#3233).
• Update packages to compile to ESM modules by default (#3112).
• Move cpu backend out of tfjs-core (#3008).

Development

• update yarn.lock (#3090).
• Refactor CI flow. [Refactor CI flow.] (#3081).

Misc

• Update tfjs-data to 2.0.0-rc.4. (#3254).
• Fix webgl bundling and test invocation
• Remove unnecessary integration tests in node and node-gpu. (#3094).

Layers (1.7.4 ==> 2.0.0-rc.4)

Features

• [core] Modularize max. (#2955).

Breaking changes

• Remove deprecated batchnormalization op (#3238).
• Update packages to compile to ESM modules by default (#3112).
• move webgl backend out of core (#3056).
• Move cpu backend out of tfjs-core (#3008).

Converter (1.7.4 ==> 2.0.0-rc.4)

Features

• added support select v2 op (#3154).

Breaking changes

• Update packages to compile to ESM modules by default (#3112).
• Move cpu backend out of tfjs-core (#3008).

Bug fixes

• allow tfjs to dep on tf-cpu latest version (#3240).
• fix tensor array gather not be able to handle indices longer than available tensors (#3157).
• remove unnecessary eager call (#3145).

Misc

• Switching to tf.io. for file access and loading savedmodels in eager model. (#3126). Thanks, @davidzats-eng.

Node (1.7.4 ==> 2.0.0-rc.4)

Features

• Add support for napi version 6. [Add support for napi version 6] (#3242).

Breaking changes

• Remove deprecated batchnormalization op (#3238).
• [node] Remove deprecated createTypeOpAttr (#3235).
• Update packages to compile to ESM modules by default (#3112).

Bug fixes

• Fix node and nodegpu build. (#3248).

Development

• Modularize conv ops. [[core] Modularize conv ops.] (#3120).
• Remove cpu backend in tfjs-node testing, use fake instead. [Remove cpu backend in tfjs-node testing, use fake instead.] (#3082).

Documentation

• correct typo on tf.node.tensorBoard doc (#3243). Thanks, @avindra.
• fix import in README.md (#3201). Thanks, @qwertyforce.

Misc

• Remove layers symlink in node-gpu. (#3031).

Wasm (1.7.4 ==> 2.0.0-rc.4)

Features

• [core] Modularize max. (#2955).
• [wasm] Add split, sqrt kernels....

V1.7.3

Core (1.7.2 ==> 1.7.3)

Misc

• Update tfjs-core to 1.7.3. (#3123).

Data (1.7.2 ==> 1.7.3)

Misc

• Update tfjs-data to 1.7.3. (#3125).

Layers (1.7.2 ==> 1.7.3)

Misc

• Update tfjs-layers, tfjs-converter to 1.7.3, 1.7.3. (#3124).

Converter (1.7.2 ==> 1.7.3)

Misc

• Update tfjs-layers, tfjs-converter to 1.7.3, 1.7.3. (#3124).

Node (1.7.2 ==> 1.7.3)

Misc

• Update tfjs-node, tfjs-node-gpu to 1.7.3, 1.7.3. (#3129).

Wasm (1.7.2 ==> 1.7.3)

Bug fixes

• Wasm fix (#3117).

Misc

• Update tfjs-backend-wasm to 1.7.3. (#3130).

v1.7.2

Core (1.7.1 ==> 1.7.2)

Features

• Modularize addN. [[core]Modularize addN.] (#3002).
• Modularize add op. [[core]Modularize add.] (#3000).
• support Op MaxPooWthArgmax in CPU and WebGL (#3010).
• [webgpu] Modularize square and squareDiff (#2962). Thanks, @axinging.
• [core] Modularize transpose op and kernel. (#2971).

Bug fixes

• fix nightly test (#3018).

Development

• Remove redundant integration tests. [Remove test-integration.] (#2970).
• [node] Clean up dependency of tfc and tfl in node tests (#2995).
• core: update the @bazel/bazel to recommended bazellisk (#2912). Thanks, @yorkie.

Misc

• Update tfjs-core to 1.7.2. (#3023).
• [core] Modularize batchnorm. (#2996).
• Fix nightly build. (#2975).
• [tfjs-core] modularize pad op

Data (1.7.1 ==> 1.7.2)

Misc

• Update tfjs-data to 1.7.2. (#3025).

Layers (1.7.1 ==> 1.7.2)

Features

• [core] Modularize transpose op and kernel. (#2971).

Documentation

• [tfjs-layers] improve doc formatting in convolutional.ts. Thanks, @ManishAradwad.

Misc

• Update tfjs-layers, tfjs-converter to 1.7.2, 1.7.2. (#3024).
• Upgrade version for integration tests. (#2990).

Converter (1.7.1 ==> 1.7.2)

Features

• support Op MaxPooWthArgmax in CPU and WebGL (#3010).

Bug fixes

• allow TF DT_UINT8 to map to tfjs int32 (#2981).

Misc

• Update tfjs-layers, tfjs-converter to 1.7.2, 1.7.2. (#3024).

Node (1.7.1 ==> 1.7.2)

Features

• support Op MaxPooWthArgmax in CPU and WebGL (#3010).

Development

• [node] Check if native addon module exists and throw error message (#3013).
• [node] Clean up dependency of tfc and tfl in node tests (#2995).
• core: update the @bazel/bazel to recommended bazellisk (#2912). Thanks, @yorkie.

Documentation

• Update README.md (#2956). Thanks, @rthadur.

Misc

• Update tfjs-node, tfjs-node-gpu to 1.7.2, 1.7.2. (#3029).

Wasm (1.7.1 ==> 1.7.2)

Features

• support Op MaxPooWthArgmax in CPU and WebGL (#3010).

Misc

• Update tfjs-backend-wasm to 1.7.2. (#3032).
• [core] Modularize batchnorm. (#2996).

v1.7.1

Core (1.7.0 ==> 1.7.1)

Features

• [core] Modularize Div, DivNoNan. (#2946).
• [WASM] Add zerosLike and onesLike kernels (#2884).

Misc

• Update tfjs-core to 1.7.1. (#2969).
• Properly propagate NIGHTLY flag. (#2965).
• [core] Move variations of the same set of tests to nightly. (#2960).
• [tfjs-core] Modularize eye and tile ops
• Remove master branch version update. (#2958).
• Fix bugs in release and publish scripts. (#2952).
• [tfjs-core] Modularize onehot
• [wasm] Test against head. (#2953).
• modularise clone op (#2935).
• Modularize random distribution tensor creation ops (#2939).
• move inputsToSave and outputsToSave to gradConfig (#2925).
• modularize broadcastTo op (#2919).
• Fix integration test, remove yalc link. (#2905).

Data (1.7.0 ==> 1.7.1)

Misc

• Update tfjs-data to 1.7.1. (#2979).
• Remove master branch version update. (#2958).
• Fix bugs in release and publish scripts. (#2952).
• Remove link-master. Refactor test steps. (#2913).
• [node]Test against head. (#2854).

Layers (1.7.0 ==> 1.7.1)

Misc

• Update tfjs-layers, tfjs-converter to 1.7.1, 1.7.1. (#2974).
• Remove master branch version update. (#2958).
• Fix bugs in release and publish scripts. (#2952).
• Remove link-master. Refactor test steps. (#2913).
• adding the default values for fit function in docs. (#2869). Thanks, @dhirensr.
• [node]Test against head. (#2854).

Converter (1.7.0 ==> 1.7.1)

Features

• Add weight shard arg to save_keras_model (#2889). Thanks, @pvaneck.

Bug fixes

• Update tf_saved_model_conversion_v2.py to allow for no model ta… (#2886). Thanks, @arvigjoka.

Misc

• Update tfjs-layers, tfjs-converter to 1.7.1, 1.7.1. (#2974).
• Remove master branch version update. (#2958).
• Fix bugs in release and publish scripts. (#2952).
• Revert "made visibility public " (#2902) (#2893).
• Remove link-master. Refactor test steps. (#2913).
• made visibility public (#2893).
• [node]Test against head. (#2854).

Node (1.7.0 ==> 1.7.1)

Misc

• Update tfjs-node, tfjs-node-gpu to 1.7.1, 1.7.1. (#2983).
• Remove master branch version update. (#2958).
• Fix bugs in release and publish scripts. (#2952).
• Remove link-master. Refactor test steps. (#2913).
• Fix integration test. (#2899).
• [node]Test against head. (#2854).

Wasm (1.5.1-alpha5 ==> 1.7.1)

Features

• [WASM] Add zerosLike and onesLike kernels (#2884).
• [wasm] Add pow, fusedMatMul. (#2788).
• [wasm] Export version. (#2819).
• Pull XNNPACK bug-fixes for WAsm SIMD (#2783).
• [wasm] Add Softmax, Neg, NotEqual. (#2728).
• Add support to NonMaxSuppressionV5 [[WASM] Add NonMaxSuppressionV5 Op] (#2598).
• [wasm] Add Gather, GatherNd kernels. (#2601).
• [wasm] Add ScatterND kernel. (#2600).

Bug fixes

• , DOC [[WASM] Update starter projects to depend on latest core and wasm] (#2706).
• [WASM] Update cdn link to point to the .js file. (#2690).

Performance

• Update XNNPACK for faster PReLU (#2787).
• Use XNNPack for Sigmoid. (#2563).

Misc

• Update tfjs-backend-wasm to 1.7.1. (#2986).
• Remove master branch version update. (#2958).
• [tfjs-core] Modularize onehot
• [wasm] Test against head. (#2953).
• Update tfjs-backend-wasm to 1.7.0. (#2864).
• Update tfjs-backend-wasm to 1.5.2-alpha1. (#2716).

v1.7.0

Core (1.6.1 ==> 1.7.0)

Features

• [wasm] Add pow, fusedMatMul. (#2788).

Misc

• Update tfjs-core to 1.7.0. (#2857).
• [tfjs-core] Refactor modular ops and kernels to new style.

Data (1.6.1 ==> 1.7.0)

Misc

• Update tfjs-data to 1.7.0. (#2860).
• Add link-to-master.js. Temporary solution for tfjs-data symlink. (#2851).

Layers (1.6.1 ==> 1.7.0)

Misc

• Update tfjs-layers, tfjs-converter to 1.7.0, 1.7.0. (#2859).
• [tfjs]Test against head. (#2852).
• Add link-to-master.js. Temporary solution for tfjs-data symlink. (#2851).
• [layers]Temporary solution to symlink dependencies. (#2842).

Converter (1.6.1 ==> 1.7.0)

Misc

• Update tfjs-layers, tfjs-converter to 1.7.0, 1.7.0. (#2859).
• fixing g3 sync error (#2855).
• [tfjs]Test against head. (#2852).
• Add link-to-master.js. Temporary solution for tfjs-data symlink. (#2851).

Node (1.6.1 ==> 1.7.0)

Misc

• Update tfjs-node, tfjs-node-gpu to 1.7.0, 1.7.0. (#2862).

v1.6.1

Core (1.6.0 ==> 1.6.1)

Bug fixes

• [core] Fix packed depthwiseConv shader to manually zero out unu… (#2838).

Misc

• Update tfjs-core to 1.6.1. (#2840).
• export a Tensor namespace to fix chained ops in g3 (#2839).

Data (1.6.0 ==> 1.6.1)

Misc

• Update tfjs-data to 1.6.1. (#2847).
• [data]Remove use of browserstack. (#2833).
• [data]Test against head. (#2832).

Layers (1.6.0 ==> 1.6.1)

Misc

• Update tfjs-layers, tfjs-converter to 1.6.1, 1.6.1. (#2841).
• [layers]Test against head. (#2831).

Converter (1.6.0 ==> 1.6.1)

Documentation

• update the README file to introduce the converter wizard CLI (#2791).

Misc

• Update tfjs-layers, tfjs-converter to 1.6.1, 1.6.1. (#2841).
• [converter]Temprory solution to symlink dependencies. (#2834).

Node (1.6.0 ==> 1.6.1)

Misc

• Update tfjs-node, tfjs-node-gpu to 1.6.1, 1.6.1. (#2853).

v1.6.0

Core (1.5.2 ==> 1.6.0)

Features

• [wasm] Add Softmax, Neg, NotEqual. (#2728).
• [core] In WebGL, if query timer extension is unavailable, retur… (#2726).

Bug fixes

• Change test-integration to look for 'run-ci' instead of 'diff'. (#2823).
• [tfjs-core] fix building exception on Windows (#2655). Thanks, @haoyunfeix.
• Fix saving at webworker (#2647). Thanks, @WenheLI.

Misc

• Update tfjs-core to 1.6.0. (#2824).
• Update tfjs-core to 1.5.3. (#2814).
• [tfjs-core] Modularize squaredDiff
• [tfjs-core] Refactor util functions (#2720). Thanks, @neutiyoo.

Data (1.5.2 ==> 1.6.0)

Documentation

• Update readers.ts (#2752). Thanks, @rthadur.

Misc

• Update tfjs-data to 1.6.0. (#2826).

Layers (1.5.2 ==> 1.6.0)

Misc

• Update tfjs-layers, tfjs-converter to 1.6.0, 1.6.0. (#2825).
• Add affected packages analysis. (#2793).
• Doc fix (#2769).

Converter (1.5.2 ==> 1.6.0)

Features

• [tfjs-converter] Allow custom shard size for more conversion pairs (#2774). Thanks, @pvaneck.

Bug fixes

• Add some fixes to converter wizard (#2799). Thanks, @pvaneck.
• fixing the tensor leak by add tidy block for sync executors and dispose intermediate tensors fro async executors (#2756).

Misc

• Update tfjs-layers, tfjs-converter to 1.6.0, 1.6.0. (#2825).
• be more specific on keras packages (#2813).
• more g3 sync fixes (#2811).
• fix g3 sync errors (#2810).
• [tfjs-converter] Adjust tfjs-converter build script (#2789). Thanks, @pvaneck.
• [converter] Test against head. (#2780).
• [converter] Remove use of browserstack (#2734).

Node (1.5.2 ==> 1.6.0)

Features

• [node] Add softmax kernel. (#2767).

Misc

• Update tfjs-node, tfjs-node-gpu to 1.6.0, 1.6.0. (#2829).
• Update package.json (#2762). Thanks, @etiennedupont.
• [tfjs-core] Modularize squaredDiff

1.5.2

Core (1.5.1 ==> 1.5.2)

Features

• Add reduce and made some changes as requested (#2617). Thanks, @NALLEIN.
• [wasm] Add Gather, GatherNd kernels. (#2601).
• [wasm] Add ScatterND kernel. (#2600).
• [webgpu] Add fusedConv2d kernel (#2448). Thanks, @axinging.

Bug fixes

• Avoid using apply/call due to new ts compiler flag (#2680).

Documentation

• Fix model save with tf.io.http and custom headers (#2696). Thanks, @rthadur.
• [Core] Throw a better error for RMSPropOptimizer when no LR (#2634).
• Fix typo in profile doc (#2171). Thanks, @Lewuathe.

Misc

• Change send email for the new monorepo nightly build trigger and also add success notification (#2639).

Converter (1.5.1 ==> 1.5.2)

Features

• support save for graph model (#2676).
• Add support to NonMaxSuppressionV5 (#2579).

Bug fixes

• Avoid using apply/call due to new ts compiler flag (#2680).

Misc

• update the tensorflow deps to 2.1.0 (#2627).

Node (1.5.1 ==> 1.5.2)

Features

• [node]Add node binding api to get number of loaded SavedModel (#2635).

Documentation

• Update saved_model.ts (#2700). Thanks, @rthadur.
• Fix doc inconsistencies in tfjs-node image (#2658). Thanks, @pvaneck.

1.5.0

Core (1.4.0 ==> 1.5.1)

Features

• [core] Add gradient registry (#2584).
• [WASM] Add Sin, Cos, Tanh, Rsqrt. (#2589).
• [WASM] Add Less, LessEqual, LogicalAnd, Tile. (#2578).
• [WASM] Add Greater, GreaterEqual kernels (#2543). Thanks, @sapphi-red.
• [WASM] Add Sum. (#2566).
• [WASM] Add Log. (#2564).
• [wasm] Add Minimum, Maximum kernels (#2528). Thanks, @sapphi-red.
• Add support to NonMaxSuppressionV5 (#2524).
• [wasm] Add Unpack kernel (#2522).
• [wasm] Add Exp tests from core (#2521).
• [wasm] Enable execution in node (#2520).

Bug fixes

• [WASM] Fix bug where we try to set more memory on WASM heap tha… (#2591).
• [core] Dispose all webgl programs when disposing the backend. (#2577).
• [WASM] Support summing across outer dimensions. (#2574).
• [Core] Bring back engine test that makes sure memory can be allocated outside a scope. (#2551).
• [wasm] Fix bug in CropAndResize, and add AutoML models to bench… (#2510).

Development

• [bench] Update benchmarks to show aggregate kernel timing or in… (#2595).
• [wasm] Fix bug in CropAndResize, and add AutoML models to bench… (#2510).
• Remove a few unnecessarily mutable exports from tfjs-core (#2498). Thanks, @blickly.
• [wasm] Add wasm and new models to the benchmark page (#2494).
• [WASM] Add readme and release script for wasm. (#2486).

Documentation

• [WASM] Update FAQ with link to github issues for feedback. (#2602).
• [wasm] Update benchmarks with Pixel 3 and Macbook Pro 15 numbers (#2573).
• Doccomment change (#2537).

Security

• Fix security errors for tfjs (#2550).

Misc

• Split NonMaxSuppressionV3 and NonMaxSuppresionV5 to two APIs (#2557).

Data (1.4.0 ==> 1.5.1)

Misc

• [data] Update microphone test wait time to remove flaky (#2548).

Converter (1.4.0 ==> 1.5.1)

Performance

• create nightly tests, cloudbuild and cloudfunc files for converter pip (#2514).

Development

• Remove unnecessarily mutable exports from tfjs-converter (#2509). Thanks, @blickly.

Misc

• Fix cloud build (#2556).
• Fix cloud build and readme (#2555).
• fix g3 lint error (#2502).

Node (1.4.0 ==> 1.5.1)

Features

• Add aarch64 support for tfjs-node (#2513). Thanks, @yhwang.
• Add NonMaxSuppressionV5 to tfjs-node (#2549).

Bug fixes

• [node] Return multiple tensor when savedModel has one input multi outputs (#2580).

Documentation

• Add aarch64 support for tfjs-node (#2513). Thanks, @yhwang.

Misc

• [node] Update libtensorflow version to 1.15.0 (#2552).