Skip to content

Releases: tensorflow/tfjs

tfjs-react-native 0.5.0

27 Oct 23:09
89c16ce
Compare
Choose a tag to compare

tfjs-react-native (0.4.0 ==> 0.5.0)

Features

  • Support loading sharded models with bundleResourceIO (#4113).

Misc

  • Update tfjs-react-native to 0.5.0. (#4145).

tfjs-react-native 0.4.0

09 Oct 15:57
1bc678c
Compare
Choose a tag to compare

tfjs-react-native (0.3.0 ==> 0.4.0)

Bug fixes

  • Update tfjs version for tfjs-react-native (#4045).
    • Fixes #3885 'func is not a function error' that occurs with version of tfjs >= 2.1.0

Misc

  • Update tfjs-react-native to 0.4.0. (#4048).

tfjs-v2.6.0

08 Oct 18:02
Compare
Choose a tag to compare

Core (2.5.0 ==> 2.6.0)

Misc

Data (2.5.0 ==> 2.6.0)

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).

Layers (2.5.0 ==> 2.6.0)

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).

Converter (2.5.0 ==> 2.6.0)

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).
  • Add Unique op (#4007). Thanks, @jinjingforever.
  • [converter] Add table initializer to graph execution. (#3975).

Node (2.5.0 ==> 2.6.0)

Bug fixes

  • Avoid keeping pointer to temporary strings (#4016). Thanks, @estyrke.

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).
  • Add Unique op (#4007). Thanks, @jinjingforever.

Wasm (2.5.0 ==> 2.6.0)

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).

Cpu (2.5.0 ==> 2.6.0)

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).
  • Fix webgl CPU forwarding for ops that have been modularized in cpu backend (#3995). Thanks, @jinjingforever.
  • Add Unique op (#4007). Thanks, @jinjingforever.

Webgl (2.5.0 ==> 2.6.0)

Misc

  • Update release branch tfjs_2.6.0 lock files.
  • Update monorepo to 2.6.0. (#4038).
  • Fix webgl CPU forwarding for ops that have been modularized in cpu backend (#3995). Thanks, @jinjingforever.
  • Add Unique op (#4007). Thanks, @jinjingforever.

tfjs-v2.5.0

07 Oct 03:03
Compare
Choose a tag to compare

Please disable CPU forwarding if you are using WebGL backend. There is a bug related to CPU forwarding in the 2.5.0 release.

tf.env().set('WEBGPU_CPU_FORWARD', false);

Core (2.4.0 ==> 2.5.0)

Features

  • support uint8 input for saved model execution (#3974).
  • [webgl] Modularize reshape, add refCounter. (#3910).

Bug fixes

  • fix Promise type backend init in Angular (#3993). Thanks, @deel77.

Documentation

  • Improve error message for missing kernels (#3916).

Security

  • update node fetch version (#3971).

Misc

  • Update monorepo to 2.5.0. (#4027).
  • Increase timeout. (#3976).
  • Add retry to karma. (#3954).
  • [cpu] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3868).

Data (2.4.0 ==> 2.5.0)

Bug fixes

Security

  • update node fetch version (#3971).

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).

Layers (2.4.0 ==> 2.5.0)

Features

Bug fixes

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).
  • [Refactoring] Clean up recurrent cells and layers getConfig (#3880). Thanks, @dikatok.
  • fix lint errors (#3955).
  • [cpu] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3868).

Converter (2.4.0 ==> 2.5.0)

Features

  • update the tf deps and moved PyInquirer to extra_requires (#3965).

Bug fixes

  • fix numpy version mismatch (#4021).
  • Fix build-npm script (#3934). Thanks, @haoyunfeix.
  • fix output tensor get disposed if output name contains index (#3920).

Security

  • update node fetch version (#3971).

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).
  • [converter]Support table initializer in converter. (#3958).
  • Semver-based upper constraints for dependencies (#4017). Thanks, @chongkong.
  • relex tfhub dep (#3992).

Node (2.4.0 ==> 2.5.0)

Features

  • support int64 dtype for saved model execution (#4008).
  • support uint8 input for saved model execution (#3974).

Bug fixes

  • support int64 dtype for saved model execution (#4008).
  • Fix build-npm script (#3934). Thanks, @haoyunfeix.

Security

  • update node fetch version (#3971).

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).
  • [tfjs-node]Support cnpm registry (#3832). Thanks, @SirM2z.

Wasm (2.4.0 ==> 2.5.0)

Bug fixes

  • [wasm] Pin rules_cc version. (#4012).

Documentation

  • Fixed typo in backend_wasm.ts (#4006). Thanks, @pabloegpf1.
  • [wasm] Fix documentation of setWasmPaths. (#3935).
  • update docs for wasm webpack setup (#3936).

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).
  • Add retry to karma. (#3954).
  • [cpu] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3868).

Cpu (2.4.0 ==> 2.5.0)

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).
  • Modularize MaxPool, MaxPoolBackprop, AvgPool, AvgPoolBackprop cpu kernels. (#3973). Thanks, @jinjingforever.
  • Modularize BatchNorm in cpu and webgl backend (#3960). Thanks, @jinjingforever.
  • Modularize the remaining unary ops in cpu backend (#3946). Thanks, @jinjingforever.
  • [cpu] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3868).
  • [cpu] Modularize NotEqual. (#3919).

Webgl (2.4.0 ==> 2.5.0)

Features

  • Check for null returned from canvas.getContext (#3970). Thanks, @jameshfisher.
  • Log error caught when creating WebGL context (#3969). Thanks, @jameshfisher.
  • [webgl] Modularize reshape, add refCounter. (#3910).

Misc

  • Update release branch tfjs_2.5.0 lock files.
  • Update monorepo to 2.5.0. (#4027).
  • [webgl] Modularize MaxPool, MaxPoolBackprop, AvgPool, AvgPoolBackprop webgl kernels (#3984). Thanks, @jinjingforever.
  • Modularize BatchNorm in cpu and webgl backend (#3960). Thanks, @jinjingforever.
  • Increase timeout. (#3963).
  • Add retry to karma. (#3954).
  • Modularize more ops in webgl backend (#3932). Thanks, @jinjingforever.
  • [cpu] Modularize cast, complex, concat, add, sub, multiply, notEqual, slice, fft, ifft. (#3868).
  • Modularize "Cos" for webgl backend. (#3921). Thanks, @jinjingforever

tfjs-v2.4.0

15 Sep 18:49
Compare
Choose a tag to compare

Core (2.3.0 ==> 2.4.0)

Features

  • Speed up toPixels. (#3813).
  • add weightUrlTranslationFunc to loadOptions (#3797).

Bug fixes

  • make tensor instanceof more robust (#3906).
  • make flags dependency in io namespace explicit (#3848).
  • add clipByValue int32 test and fix the node impl (#3822).

Documentation

Misc

  • Distinguish ops from kernels in tf.profile (#3897).

Data (2.3.0 ==> 2.4.0)

Documentation

Layers (2.3.0 ==> 2.4.0)

Features

Bug fixes

  • add functional model to support changes from TF 2.3 (#3863).

Documentation

Converter (2.3.0 ==> 2.4.0)

Bug fixes

  • fix tensorarray disposing weight tensors (#3847).
  • remove code that specifying the detail of strip_unused_nodes (#3836).
  • Update keras_h5_conversion_test.py (#3794). Thanks, @rthadur.

Development

  • remove chain api use from converter (#3824).

Documentation

Misc

  • fix converter node build (#3878).

Node (2.3.0 ==> 2.4.0)

Bug fixes

  • add clipByValue int32 test and fix the node impl (#3822).

Documentation

Wasm (2.3.0 ==> 2.4.0)

Features

Bug fixes

  • [wasm] Exclude node from the threaded version. (#3829).
  • add clipByValue int32 test and fix the node impl (#3822).

Documentation

Misc

  • [wasm] Add more device testing (#3851).

Cpu (2.3.0 ==> 2.4.0)

Bug fixes

  • add clipByValue int32 test and fix the node impl (#3822).

Misc

  • Add Cos to kernel registry (cpu backend) (#3907).
  • [cpu]Use flat array instead of nested array in dilation2d kernel (#3860).
  • Modularize identity kernel. (#3859).
  • Implement cpu refCounter, Modularize kernels Reshape and SpaceToBatchND (#3659).

Webgl (2.3.0 ==> 2.4.0)

Features

  • Speed up toPixels. (#3813).

Bug fixes

  • Fix memleak in modularized reduce. (#3872).

Documentation

tfjs-v2.3.0

14 Aug 07:31
Compare
Choose a tag to compare

Core (2.2.0 ==> 2.3.0)

Bug fixes

  • move setOpHandler to side_effects (#3800).

Misc

  • Update monorepo to 2.3.0. (#3802).

Data (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

Layers (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

Converter (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

Node (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

Wasm (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

Cpu (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

Webgl (2.2.0 ==> 2.3.0)

Misc

  • Update release branch tfjs_2.3.0 lock files.
  • Update monorepo to 2.3.0. (#3802).

tfjs-v2.2.0

13 Aug 23:53
ef9b917
Compare
Choose a tag to compare

Core (2.1.0 ==> 2.2.0)

Features

  • [core] Add tf.image.flipLeftRight (#3779).
  • Add kernel timing information to tf.profile() (#3721). Thanks, @Linchenn.

Bug fixes

Development

  • Remove chained ops usage from core (#3706).
  • custom bundle tool (#3692).

Documentation

Misc

  • Update monorepo to 2.2.0. (#3796).

Data (2.1.0 ==> 2.2.0)

Misc

  • Update monorepo to 2.2.0. (#3796).

Layers (2.1.0 ==> 2.2.0)

Misc

  • Update monorepo to 2.2.0. (#3796).

Converter (2.1.0 ==> 2.2.0)

Features

  • Add experiments flag to converter (#3748).

Bug fixes

  • ensure clone is done only for not kept tensors (#3742).

Development

  • Add TF-Op to tfjs-op mapper to converter (#3726).

Misc

  • Update monorepo to 2.2.0. (#3796).

Node (2.1.0 ==> 2.2.0)

Features

  • [core] Add tf.image.flipLeftRight (#3779).

Misc

  • Update monorepo to 2.2.0. (#3796).

Wasm (2.1.0 ==> 2.2.0)

Features

  • [core] Add tf.image.flipLeftRight (#3779).
  • [wasm] Make setWasmPaths require an entry for each WASM binary. (#3749).
  • [wasm] Add support for pthreads. (#3627).

Bug fixes

  • [wasm] Fix logic for getting path to WASM binary. (#3764).

Development

  • Organize the benchmark toolkit (#3793). Thanks, @Linchenn.
  • Add wasm backend to custom build tool (#3750).
  • [wasm] Fix worker module generation script. (#3743).

Misc

  • Update monorepo to 2.2.0. (#3796).

tfjs-v2.1.0

04 Aug 17:14
0230521
Compare
Choose a tag to compare

Core (2.0.1 ==> 2.1.0)

Features

  • Decouple logging from profiler.profileKernel (#3606). Thanks, @Linchenn.
  • Change image rotation kernel to be counterclockwise. (#3626).
  • Modularize spectral ops. (#3608).
  • Support async query for tf.profile (#3564). Thanks, @Linchenn.
  • Add tf.image.rotateWithOffset op. (#3560).
  • add various op fixes for ddsp models (#3539).
  • [wasm] Add Reverse kernel. (#3526).
  • [core] Remove reshape from tensor / opHandler. (#3514).
  • Modularize stack, unstack, reshape, expandDims, squeeze (#3452).
  • [wasm] Add Conv2dBackpropInput kernel. (#3472).
  • [WASM] Add OneHot op (#3409). Thanks, @pvaneck.
  • [WASM] Add equal op (#3456). Thanks, @pvaneck.

Bug fixes

  • [core] Account for optional inputs in profileKernel. (#3697).
  • [wasm] Pass in fusedDepthwiseConv flag when computing conv_info. (#3676).
  • fix import path in kernel_names (#3653).
  • fix cumsum and strided slice bugs (#3638).
  • add various op fixes for ddsp models (#3539).
  • silence warning about overriding gradients (#3572).
  • Fix min op test for nightly [Fix min op test for nightly] (#3547).
  • Attempt to fix nightly (#3545).

Development

  • Modularize fused ops (#3597).
  • Modularize remaining unary ops (#3605).
  • Modularize softmax/logSoftmax, sparseToDense, TopK, stridedSlice (#3587).
  • Modularize tensor ops (#3590).
  • Modularize array ops (#3588).
  • modularize unary ops (#3586).
  • modularize signal ops (#3579).
  • Modularize slice ops (#3581).
  • Modularise unary ops (#3575).
  • silence warning about overriding gradients (#3572).
  • Modularize sign, ceil, floor, neg (#3569).
  • Modularize scatterNd, unsortedSegmentSum & gather. (#3544).
  • Move all the remaining chaining api outside of tensor.ts (#3538).
  • Modularize argMax, argMin, min, mean and sum (#3517).
  • Modularize any, all, prod, moments and logSumExp (#3505).
  • modularize reverse ops (#3502).
  • modularize loss ops (#3492).
  • remove chaining api from movingAverage op (#3499).
  • modularize lstm ops (#3498).
  • remove chaining api from norm (#3500).
  • Modularize cropAndResize (#3474).
  • Modularise logical ops (#3465).
  • modularise linalg ops (#3464).
  • Modularize nonMaxSuppression ops (#3461).
  • Modularise resizeBilinear and resizeNearestNeighbor (#3430).

Documentation

  • added docstring for compute and apply gradients (#2926). Thanks, @dhirensr.
  • Update License (#3507).
  • Update License (#3507).

Misc

  • Update monorepo to 2.1.0. (#3730).
  • Add NonMaxSuppressionV4. (#3602).
  • Remove debug from arg list. (#3530).
  • [core] Account for both ellipsisMask and begin / end masks in strided slice. (#3488).
  • Implement op dilation2d (#3462).

Data (2.0.1 ==> 2.1.0)

Documentation

Misc

  • Update monorepo to 2.1.0. (#3730).

Layers (2.0.1 ==> 2.1.0)

Documentation

Misc

  • Update monorepo to 2.1.0. (#3730).

Converter (2.0.1 ==> 2.1.0)

Features

  • add various op fixes for ddsp models (#3539).
  • Improve model summary (#3519).
  • added support explicit padding for conv2d ops (#3493).
  • Add Support for various TensorList ops (#3432).

Bug fixes

  • fix model conversion error due to TF 2.3.0 release difference (#3685).
  • fix cumsum and strided slice bugs (#3638).
  • add various op fixes for ddsp models (#3539).
  • fix functions ops summary (#3506).
  • support negative split value (#3482).

Documentation

Misc

  • Update monorepo to 2.1.0. (#3730).
  • Add NonMaxSuppressionV4 to converter. (#3616).
  • upgrade pylint version for python3 and fix lints (#3556).
  • update the python requirements.txt to set version limit properly (#3523).
  • Implement op dilation2d (#3462).
  • update the model summary tool to check both sub functions (#3481).
  • maintain single id tensor for each tensor list and tensor array (#3479).

Node (2.0.1 ==> 2.1.0)

Features

  • Add tf.image.rotateWithOffset op. (#3560).

Documentation

Misc

  • Update monorepo to 2.1.0. (#3730).
  • [tfjs-node] Fix bug in node.tensorBoard() callback re initialEpoch (#3714).
  • Add NonMaxSuppressionV4. (#3602).
  • Implement op dilation2d (#3462).
  • Move kernel registration to one file. (#3476).

Wasm (2.0.1 ==> 2.1.0)

Features

  • [wasm] Modularize remaining kernels. (#3663).
  • Change image rotation kernel to be counterclockwise. (#3626).
  • [wasm] Use XNNPACK for PadV2. (#3604).
  • [wasm] Add XNNPACK operators for unary ops (Square, Neg, Abs). (#3598).
  • Add tf.image.rotateWithOffset op. (#3560).
  • [wasm] Use export KernelConfig pattern in WASM. (#3571).
  • add various op fixes for ddsp models (#3539).
  • [wasm] Use clone in transpose. (#3570).
  • [wasm] Add Reverse kernel. (#3526).
  • [core] Remove reshape from tensor / opHandler. (#3514).
  • Modularize stack, unstack, reshape, expandDims, squeeze (#3452).
  • [WASM] Use XNNPack for div (#3480). Thanks, @pvaneck.
  • [wasm] Add Conv2dBa...
Read more

tfjs-vis-v1.4.3

17 Jun 14:05
57ef0d0
Compare
Choose a tag to compare

tfjs-vis (1.1.0 ==> 1.4.3)

We missed updating release notes for a recent point releases, so these notes summarize changes between 1.1.0 ==> 1.4.3. We'll also be using GitHub release notes over the CHANGELOG.md file in tfjs-vis.

Features

  • [tfjs-vis] support custom colors in confusion matrix
  • [tfjs-vis] Allow custom mark colors in charts
  • [tfjs-vis] Support custom tickLabels with duplicate values in heatmaps (#2012).

Bug fixes

  • [tfjs-vis] update vega dependency (#3449).
  • Upgrade vega versions in tfjs-vis (#3428).
  • [tfjs-vis] Upgrade vega and fix tooltip displays (#1980).
  • update barchart & histrogram tooltips (#2070).

Documentation

  • [tfjs-vis] update readme, add more screenshots (#2914).

Misc

  • [tfjs-vis] remove glamor-tachyons

TensorFlow.js 2.0.1

16 Jun 12:58
44994f0
Compare
Choose a tag to compare

Core (2.0.0 ==> 2.0.1)

Features

  • [core] Support explicit padding in tf.conv. (#3427).
  • [wasm] Add SIMD detection. (#3344).
  • [wasm] Add where / select. (#3376).
  • [core] Modularize tf.fill and add to WASM backend. (#3388).
  • float16 quantization + granular quantization (#3340). Thanks, @JesseFarebro.
  • Support Complex64 weights conversion and loading (#3362).
  • Upgrade Emscripten and add correctness test to benchmarks UI. (#3342).
  • Modularize cumsum. (#3296).
  • Add support for ellipsisMask in stridedSlice. (#3304).

Bug fixes

  • Fix saving to IndexedDB inside web worker (#3403). Thanks, @paulkre.
  • [wasm] Ensure concatenation axes are parsed. (#3387).

Development

  • Modularize relu6, leakyRelu, prelu, selu. [Modularize relu6, leakyRelu, prelu, selu.] (#3431).
  • Modularise atan2, floorDiv, mod and mul (#3407).
  • modularize GatherNd (#3393).
  • Modularize minimum & maximum (#3379).
  • align fromPixels op with kernel calling convention (#3391).
  • modularize complex ops (#3392).
  • modularize pow op (#3369).
  • Modularize relu. [Modularize relu.] (#3336).
  • Fix maxPool save. [Fix maxPool save.] (#3337).
  • Modularize pool ops. [Modularize pool ops.] (#3269).

Documentation

  • Add docs about backends to tfjs-core readme (#3382). Thanks, @pvaneck.
  • Add op modularization guide. (#3305).

Misc

Data (2.0.0 ==> 2.0.1)

Misc

  • Update tfjs-data to 2.0.1. (#3448).

Layers (2.0.0 ==> 2.0.1)

Features

  • fixed tf 2.2.0 weight error (#3370).

Bug fixes

  • reenable the GRU integration test, fixed the shape mismatch error (#3377).

Documentation

Misc

  • Update tfjs-layers, tfjs-converter to 2.0.1, 2.0.1. (#3443).
  • Clean up integration test in layers. (#3338).

Converter (2.0.0 ==> 2.0.1)

Features

  • add support for tensor list (#3411).
  • add stateful while and if ops (#3395).
  • float16 quantization + granular quantization (#3340). Thanks, @JesseFarebro.
  • added Cumsum and broadcastTo to converter (#3390).
  • Adds loadSync to GraphModel (#3383). Thanks, @davlhd.
  • Fix issue with control flow v2 nested loop (#3380).
  • fixed tf 2.2.0 weight error (#3370).
  • Support Complex64 weights conversion and loading (#3362).
  • Adding support for Control flow v2 in converter and executor (#3346).

Documentation

Misc

  • Update tfjs-layers, tfjs-converter to 2.0.1, 2.0.1. (#3443).

Node (2.0.0 ==> 2.0.1)

Features

  • [core] Support explicit padding in tf.conv. (#3427).

Misc

  • Update tfjs-node, tfjs-node-gpu to 2.0.1, 2.0.1. (#3453).

Wasm (2.0.0 ==> 2.0.1)

Features

  • [wasm] Enable SIMD. (#3355).
  • [WASM] Add CommonJS output in rollup (#3378). Thanks, @pvaneck.
  • [wasm] Add transpose support to Max kernel. (#3414).
  • [wasm] Add SIMD detection. (#3344).
  • [wasm] Upgrade XNNPACK and its dependencies. (#3343).
  • [wasm] Add where / select. (#3376).
  • [core] Modularize tf.fill and add to WASM backend. (#3388).
  • Upgrade Emscripten and add correctness test to benchmarks UI. (#3342).

Bug fixes

  • [wasm] Ensure concatenation axes are parsed. (#3387).
  • [WASM] Fix allocation issue for buffers with byteOffsets (#3334). Thanks, @pvaneck.

Development

  • Modularise atan2, floorDiv, mod and mul (#3407).
  • modularize GatherNd (#3393).
  • Modularize pool ops. [Modularize pool ops.] (#3269).

Misc

  • Update tfjs-backend-wasm to 2.0.1. (#3454).