Skip to content

TFJS 4.3.0
Compare
Choose a tag to compare
@mattsoulanille mattsoulanille released this 20 Mar 19:37
· 254 commits to master since this release
tfjs-v4.3.0
a8a3f52
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Update 2023-03-21: This Release is Deprecated

This release contains a regression that causes some graph models to dispose their tensors too early. See #7504 for details.

Core (4.2.0 ==> 4.3.0)

Features

  • Update flags.ts (#7367). Thanks, @Linchenn.
  • [op] added support for TensorScatterUpdate op (#7189).

Bug fixes

Performance

Documentation

Misc

  • Update monorepo to 4.3.0. (#7488).
  • [wasm] Add Multinomial kernel (#7468). Thanks, @chunnienc.
  • Fix gather signature in chained ops (#7465).
  • disable offscreen canvas for safari, since gl fencing is not working yet for its technology preview (#7436).
  • Propagate errors from indexedDB database.put (#7322).
  • Add StaticRegexReplace Op (#7379).
  • SharedArrayBuffer is not GPU data (#7344). Thanks, @axinging.
  • Support MatrixBandPart and allow numLower to be a tensor (#7351). Thanks, @Linchenn.
  • Add a case for atanh when input is 1 (#6964). Thanks, @xhcao.
  • [tfjs-node] support einsum in tfjs node (#7321).
  • fix comment in tile.ts (#7298). Thanks, @sonmh79.

Data (4.2.0 ==> 4.3.0)

Misc

  • Update monorepo to 4.3.0. (#7488).
  • build(deps): bump vega in /tfjs-data/demo/boston-housing (#7444). Thanks, @dependabot[bot].
  • build(deps): bump vega-functions in /tfjs-data/demo/boston-housing (#7442). Thanks, @dependabot[bot].
  • build(deps): bump json5 in /tfjs-data/demo/boston-housing (#7219). Thanks, @dependabot[bot].
  • build(deps): bump fast-json-patch in /tfjs-data/demo/boston-housing (#7211). Thanks, @dependabot[bot].

Layers (4.2.0 ==> 4.3.0)

Misc

  • Update monorepo to 4.3.0. (#7488).
  • Typo fix (#7393). Thanks, @dmotz.
  • #7376 Allow seeding the Orthogonal Initializer and use QR instead of gramSchmidt (#7377). Thanks, @cosminc98.

Converter (4.2.0 ==> 4.3.0)

Features

  • [converter] add TensorScatterUpdate op to converter and executor (#7246).

Bug fixes

  • allow weight only keras H5 file that lacks keras version info to be parsed without error (#7359).

Misc

  • Update tfjs_4.3.0 lockfiles (#7492).
  • Update monorepo to 4.3.0. (#7488).
  • Never pass a NoOp value to an op (#7460).
  • Fix topological sort for sync nodes execution order (#7431). Thanks, @chunnienc.
  • Remove array.at (#7409). Thanks, @chunnienc.
  • Add new tensor diposal algorithm for sync graph execution (#7392). Thanks, @chunnienc.
  • Add cache for parseNodeName (#7391). Thanks, @chunnienc.
  • Update rules_python to 0.18.1 (#7383).
  • Add StaticRegexReplace Op (#7379).
  • added converter support for inf and infinite ops (#7365).
  • Support MatrixBandPart and allow numLower to be a tensor (#7351). Thanks, @Linchenn.
  • Support tfjs converter options in JAX conversion (#7331).
  • Update TypeScript to 4.9.4 (#7300).
  • Debug converter python code with debugpy (#7283).
  • Refactor python converter functions to not call each other (#7281).
  • Rename bazel BUILD files (#7253). Thanks, @chunnienc.
  • Make converter hashtable test not overly specific (#7249).
  • build(deps): bump json5 from 2.1.0 to 2.2.3 in /tfjs-converter/demo (#7240). Thanks, @dependabot[bot].

Node (4.2.0 ==> 4.3.0)

Features

  • [op] added support for TensorScatterUpdate op (#7189).

Misc

  • Update tfjs_4.3.0 lockfiles (#7492).
  • Update monorepo to 4.3.0. (#7488).
  • [wasm] Add Multinomial kernel (#7468). Thanks, @chunnienc.
  • Add StaticRegexReplace Op (#7379).
  • [tfjs-node] support einsum in tfjs node (#7321).
  • Update TypeScript to 4.9.4 (#7300).
  • Improve npm publishing infra (#7287).
  • build(deps): bump json5 from 2.2.0 to 2.2.3 in /tfjs-node-gpu (#7243). Thanks, @dependabot[bot].
  • build(deps): bump json5 from 2.2.0 to 2.2.3 in /tfjs-node (#7242). Thanks, @dependabot[bot].
  • Do not run docker interactively in tfjs-node publish (#7227).

Wasm (4.2.0 ==> 4.3.0)

Features

  • [wasm] Add atan kernel (#7252). Thanks, @chunnienc.
  • [op] added support for TensorScatterUpdate op (#7189).

Security

  • build(deps): bump json5 in /tfjs-backend-wasm/starter/webpack (#7218). Thanks, @dependabot[bot].

Misc

Cpu (4.2.0 ==> 4.3.0)

Features

  • [op] added support for TensorScatterUpdate op (#7189).

Bug fixes

Misc

  • Update monorepo to 4.3.0. (#7488).
  • Add unique (#7469). Thanks, @chunnienc.
  • Add StaticRegexReplace Op (#7379).
  • [WebGPU] Support forwarding FloorDiv to CPU (#7336).

Webgl (4.2.0 ==> 4.3.0)

Features

  • [WebGL] Implement packed ScatterND (#7292). Thanks, @Linchenn.
  • [op] added support for TensorScatterUpdate op (#7189).

Bug fixes

Performance

Misc

  • Update tfjs_4.3.0 lockfiles (#7492).
  • Update monorepo to 4.3.0. (#7488).
  • React native update tfjs to 4.2.0 (#7451).
  • disable offscreen canvas for safari, since gl fencing is not working yet for its technology preview (#7436).
  • [webgl] Reduce JS map lookups in webgl backend (#7387). Thanks, @chunnienc.
  • webgl: Add packed op for sin/cos (#7382). Thanks, @qjia7.
  • Add StaticRegexReplace Op (#7379).
  • Fix license headers added in 2023 (#7353). Thanks, @chunnienc.

Webgpu (0.0.1-alpha.17 ==> 0.0.1-alpha.18)

Misc

  • Update monorepo to 4.3.0. (#7488).
  • webgpu: Remove the unnecessary conditions (#7470). Thanks, @qjia7.
  • webgpu: Optimize broadcast mul/div (#7437). Thanks, @qjia7.
  • webgpu: Switch back to old karma reporter (#7449). Thanks, @gyagp.
  • [WebGPU] support AvgPool3DGrad kernel (#7440). Thanks, @xhcao.
  • [WebGPU] support conv3d backprop kernels (#7434). Thanks, @xhcao.
  • [webgpu] Add karma-jasmine-html-reporter to show test status in the b… (#6636). Thanks, @haoyunfeix.
  • webgpu: Support any component buffer (#7426). Thanks, @qjia7.
  • webgpu: Fix conv2d gradient cases failures (#7433). Thanks, @qjia7.
  • [WebGPU] support pool3d kernels (#7405). Thanks, @xhcao.
  • webgpu: Optimize Softmax (#7411). Thanks, @qjia7.
  • [WebGPU] Support ResizeBilinearGrad kernel (#7385). Thanks, @xhcao.
  • webgpu: Skip staticRegexReplace test cases (#7390). Thanks, @gyagp.
  • webgpu: Support 2x4 output for Conv2DBackpropInput (#7371). Thanks, @qjia7.
  • [WebGPU] support ResizeNearestNeighborGrad kernel (#7354). Thanks, @xhcao.
  • webgpu: NaN refactor (#7352). Thanks, @qjia7.
  • [WebGPU] Support forwarding FloorDiv to CPU (#7336).
  • build(deps): bump ua-parser-js in /tfjs-backend-webgpu (#7312). Thanks, @dependabot[bot].
  • webgpu: unify the uniform variables' names of conv (#7279). Thanks, @xhcao.
  • webgpu: enable more test cases (#7278). Thanks, @xhcao.
  • webgpu: support LRNGrad kernel (#7196). Thanks, @xhcao.
  • webgpu: support BroadcastArgs kernel (#7247). Thanks, @xhcao.
  • webgpu: support MaxPoolGrad kernel (#7261). Thanks, @xhcao.
  • webgpu: add a naive implementation of conv3d (#7016). Thanks, @xhcao.
  • build(deps): bump json5 from 2.2.0 to 2.2.3 in /tfjs-backend-webgpu (#7244). Thanks, @dependabot[bot].
  • webgpu: support maxPoolWithArgmax kernel (#7176). Thanks, @xhcao.
  • webgpu: support EluGrad kernel (#7197). Thanks, @xhcao.
  • [webgpu] Skip tensorScatterUpdate unit test (#7255). Thanks, @hujiajie.
  • [webgpu] Let vectorized binary op return an overridable value on NaN (#7198). Thanks, @hujiajie.
  • webgpu: support DepthwiseConv2dNativeBackprop kernels (#7232). Thanks, @xhcao.
  • webgpu: Remove isVectorA check (#7233). Thanks, @qjia7.
  • webgpu: support multinomial operator (#7154). Thanks, @xhcao.

Contributors

gyagp, dmotz, and 11 other contributors
Assets 2
Loading