profile
viewpoint
Daniel Smilkov dsmilkov Google Cambridge, MA smilkov.com Google Brain. Co-author of TensorFlow.js. MIT Media Lab alum.

alexandersimoes/d3plus 1505

A javascript library that extends D3.js to enable fast and beautiful visualizations.

PAIR-code/saliency 505

TensorFlow implementation for SmoothGrad, Grad-CAM, Guided backprop, Integrated Gradients and other saliency techniques

google/XNNPACK 324

High-efficiency floating-point neural network inference operators for mobile, server, and Web

PAIR-code/umap-js 112

JavaScript implementation of UMAP

PAIR-code/federated-learning 111

Federated learning experiment using TensorFlow.js

conversationai/conversationai-models 102

A repository to house model building experiments and tools that are part of the Conversation AI effort.

PAIR-code/scatter-gl 13

Interactive 3D / 2D webgl-accelerated scatter plot point renderer

philsalesses/Place-Pulse 13

Place Pulse code repository

startedjasonmayes/Real-Time-Person-Removal

started time in 29 minutes

issue commenttensorflow/tfjs

Session fail to run with error: indices = 1

  • tf.tensor(['sdfsdf']) infers shape of [1]
  • tf.tensor(['sdfsdf'], [1, 1], 'string') gives shape of [1, 1].

Both are working as intended. If you want to infer shape of [1,1], you have to do tf.tensor([['sdfsdf']])

Nithanaroy

comment created time in 5 days

issue commenttensorflow/tfjs

Long-running shaders result in webgl context lost

Thanks! We won't be able to fix this for several reasons:

  • The fix for this is non-trivial and likely impossible without breaking the API since we would need to "release the GPU and the main thread", but we probably won't be able to keep the entire matmul operation synchronous (tf.matmul().dataSync() will break).
  • Huge matmuls like this do not show up in practical ML models, thus making this a low priority for the time being.
decentralion

comment created time in 5 days

push eventtensorflow/tfjs

Sandeep N Gupta

commit sha 90f52ddb341d5079925af81000edfe845e40774a

Added Handtrack.js to Gallery.md (#2725) DOC

view details

push time in 15 days

PR merged tensorflow/tfjs

Reviewers
Added Handtrack.js to Gallery.md cla: yes

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+1 -0

1 comment

1 changed file

sandeepngupta

pr closed time in 15 days

pull request commenttensorflow/tfjs

Added Handtrack.js to Gallery.md

Thanks @sandeepngupta!! :)

sandeepngupta

comment created time in 15 days

push eventtensorflow/tfjs

WenheLI

commit sha 5afa7c289e5197f141161c1f5ee85d68694c1e34

Fix saving at webworker (#2647) BUG This PR is to fix the bug while saving models at webworker env. (https://github.com/tensorflow/tfjs/issues/2643) Note: Under webworker, we can only save models via IndexedDB. Maybe we should mention it at the docs?

view details

push time in 15 days

PR merged tensorflow/tfjs

Fix saving at webworker cla: yes

This PR is to fix the bug while saving models at webworker env. (https://github.com/tensorflow/tfjs/issues/2643)

Note: Under webworker, we can only save models via IndexedDB. Maybe we should mention it at the docs?

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+5 -2

0 comment

2 changed files

WenheLI

pr closed time in 15 days

push eventtensorflow/tfjs

Neuti Yoo

commit sha fdd4603474f001c2f0befe6bdc7f5cd56c9ac24e

[tfjs-core] Refactor util functions (#2720) INTERNAL This PR refactors the following functions: - `isTensorInList` (Array.some) - `arrayBufferToBase64String` (Array.reduce) By making these declarative, in my opinion, we can reduce the cognitive load and manage the codes better.

view details

push time in 15 days

PR merged tensorflow/tfjs

[tfjs-core] Refactor util functions cla: yes

This PR refactors the following functions:

  • isTensorInList (Array.some)
  • arrayBufferToBase64String (Array.reduce)

By making these declarative, in my opinion, we can reduce the cognitive load and manage the codes better.

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+1 -6

5 comments

1 changed file

neutiyoo

pr closed time in 15 days

issue commenttensorflow/tfjs

tf.time() kernelMs returns zero in Firefox

Talked to Nikhil. We decided that we shouldn't silently ignore (omit information) b/c that is not helpful to the user. E.g. if I'm using firefox + webgl and call tf.time, I want to understand why I can't measure the kernel time.

Here is the proposal. kernelMs maps to:

  • if everything works, a number
  • if we can't measure:
{error: 'WebGL query timers are not supported in this environment'}
MatPoliquin

comment created time in 15 days

issue commenttensorflow/tfjs

tf.time() kernelMs returns zero in Firefox

Instead of omitting (which is basically silencing the error), how about we throw an error with a user-friendly msg? Specifically the webgl's backend.time() should throw an error if the query timer is not supported, and that error should propagate through tf.time. WDYT?

MatPoliquin

comment created time in 15 days

push eventWenheLI/tfjs

Ping Yu

commit sha 642a5419195de3747ed696757c6907f5ae188cd6

update the tensorflow deps to 2.1.0 (#2627) INTERNAL * update the tensorflow deps to 2.0.0 * fix lint * use 2.1.0rc2 due to an error on error saved model api * fix integration tests * fix pylint error and clean up the pip test * fix the pip package test * relex the pip requirement for h5py * update to 2.1.0

view details

Paul Van Eck

commit sha d0988eedf7bee90fb597e86f0231783e14f70565

Fix doc inconsistencies in tfjs-node image (#2658) DOC

view details

Yannick Assogba

commit sha 695f4aa6f8a6d259cc10e8901ca617dcaacad1dc

[tfjs-react-native] Add functions for streaming camera data into tensors. Adds functions for converting camera data to tensors for realtime use in model predictions. Primary goal of this change is to enable real time use of camera data with models like posenet/bodypix etc.

view details

Daniel Smilkov

commit sha cb7ed0788cdb6880d2afd1ed7d5ec9008a3c153c

Update main README.md with all the packages in the mono repo (#2604) DOC

view details

Yannick Assogba

commit sha 6a9cd1512422bf89fc46c6a3877cf8e193e5cb0d

[tfjs-vis] confusion matrix custom colors FEATURE Support custom colors in confusion matrix BUG Fix xlabel and ylabel setting in confusion matrix

view details

Yannick Assogba

commit sha 93e0e2aa675bbed5fec340441c4c2cb93ec659ff

Update tfjs-vis to 1.4.0. INTERNAL

view details

Ping Yu

commit sha 1e9faf84523d91a2516bec5703b985ab0dde09dd

support save for graph model (#2676) FEATURE * support save for graph model * fix lint

view details

Yannick Assogba

commit sha e16d9402c9a4da9dc4aed045bced3201bed4d460

Add reminder to release website to release script INTERNAL

view details

Yannick Assogba

commit sha 2b30aaa136a37c5fea2facb1253b09f02ed8d8b6

tfjs-react-native] add helper component for camera with tensor stream FEATURE

view details

Xu Xing

commit sha a5b38ed98068f3d865cba482e51accace346b8ab

[webgpu] Add numBytesAllocated (#2526) FEATURE Tensor.dispose will not free the GPU's video memory, but push the buffer into a freeBuffers Map. The freeBuffers is TFJS's free buffer, but to the GPU, it is allocated buffer (Suppose it is virtual memory). If not call Tensor.dispose in loop, the freeBuffers will grow bigger as the size of the loop get bigger. And finally the GPU run out of memory. At the same time, numBytesInGPU keeps growing. If call Tensor.dispose in loop, numBytesInGPU is very small(4194304), and remains unchanged during all trials. However, the allocated memory maybe much more bigger(847249424) than numBytesInGPU. The application may run out of memory when the numBytesInGPU keeps small. So numBytesInGPU is not a good indicator for GPU memory usage. In this CL, introduced numBytesAllocatedInGPU. numBytesAllocatedInGPU is a indicator to measure how much memory is allocated from GPU's video memory. It difers from numBytesInGPU in that: numBytesInGPU indicates how much memory is used by Tensors. numBytesAllocatedInGPU indicates the TFJS application has allocated how much memory. Some of the memory is used by Tensors, some is just kept in the freeBuffers pool. For the OOM case, it is a better indicator. numBytesInGPU = memory of accquireBuffer - memory of releaseBuffer. numBytesAllocatedInGPU = memory of GPUDevice.createBuffer - memory of GPUBuffer.destroy Hints for how to use benchmark's time function: choose a trials and reps carefully, too big trials and reps may lead to unstable performance data or OOM. Test result: 1. Not call dispose in loop, disposeAfterEachTrial === false Test case: it('transpose', async () => { const x = tf.randomNormal([1024, 1024]); await time(() => tf.transpose(x, [1, 0]), null, false, 10, 200); }); Memory usage data: numBytesInGPU=838860800, numBytesAllocatedInGPU=847249424 numBytesInGPU=1673527296, numBytesAllocatedInGPU=1681915920 numBytesInGPU=2508193792, numBytesAllocatedInGPU=2516582416 numBytesInGPU=3342860288, numBytesAllocatedInGPU=3351248912 numBytesInGPU=4177526784, numBytesAllocatedInGPU=4185915408 numBytesInGPU=5012193280, numBytesAllocatedInGPU=5020581904 numBytesInGPU=5846859776, numBytesAllocatedInGPU=5855248400 numBytesInGPU=6681526272, numBytesAllocatedInGPU=6689914896 numBytesInGPU=7516192768, numBytesAllocatedInGPU=7524581392 numBytesInGPU=8350859264, numBytesAllocatedInGPU=8359247888 2. Call dispose in loop, disposeAfterEachTrial===true Test case: it('transpose', async () => { const x = tf.randomNormal([1024, 1024]); await time(() => tf.transpose(x, [1, 0]), null, true, 10, 200); }); Memory usage data: numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 numBytesInGPU=4194304, numBytesAllocatedInGPU=847249424 Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Ruikun Cao

commit sha e6cf3a53663fe9af998c1ddb1254264b08a9bd11

add pipcook to gallery (#2686) DOC * add pipcook to gallery * add pipcook to gallery * Update Gallery to indicate TFX

view details

Yannick Assogba

commit sha 5092055115cdae4e3a94e99c64c2afea66233900

[tfjs-react-native] update api docs for react native DOC

view details

Yannick Assogba

commit sha 978d0c764bc1a079f515c3f45028fa6bce5b343e

Add tfjs-react-native package to release script INTERNAL

view details

Yannick Assogba

commit sha 0555f5726c356e9eae78bb406f75ef77c36af0ff

Update tfjs-react-native to 0.2 INTERNAL * Update tfjs-react-native to 0.2.

view details

Yannick Assogba

commit sha 8c83352385fd16a982f72a8a8e23f2a6fdc5eab6

tfjs-react-native fix version (add patch indicator) INTERNAL

view details

Yannick Assogba

commit sha 74591b8991fcdf39a31205df890a6dd5a86ffa07

[tfjs-react-native] include tests on npm. Bump version to 0.2.1 INTERNAL

view details

Nikhil Thorat

commit sha 931e0c10306722c2e403cf8cc04cc13c0e68da04

[WASM] Update cdn link to point to the .js file. (#2690) BUG Fixes https://github.com/tensorflow/tfjs/issues/2683 Without this the CDN 404s because it tries to fetch the .wasm path relative to the base jsdelivr link (jsdelivr does not redirect). The alternative is we use tf.setWasmPath here.

view details

Na Li

commit sha 78add066f2d6cf2c3bc51d7996fda5186b3126dc

Avoid using apply/call due to new ts compiler flag (#2680) BUG * Avoid using apply/call due to new ts compiler flag * Solve browserstack timeout.

view details

Rajeshwar Reddy T

commit sha a60f32fbb3d12801c35e464a1f5fd173d575097a

Fix model save with tf.io.http and custom headers (#2696) DOC

view details

Rajeshwar Reddy T

commit sha ac52f1c9e4b6319dcfaa79b685069c91be95cfde

Update saved_model.ts (#2700) DOC * Update saved_model.ts * Update saved_model.ts * Update saved_model.ts * Update saved_model.ts * Update saved_model.ts

view details

push time in 18 days

issue closedtensorflow/tfjs

The name tf.CustomCallbackConfig is not descriptive enough

In tfjs, tf.CustomCallbackConfig comes from tfjs-layers. But the name doesn't reflect that fact and it's unclear what the config is for (it is for tf.Model.fit). It should be moved to a different name and/or different namespace to be clearer.

closed time in 18 days

caisq

issue commenttensorflow/tfjs

The name tf.CustomCallbackConfig is not descriptive enough

With the improved doc generator, we now generate much better info about the possible callback fields, thus closing this issue.

caisq

comment created time in 18 days

issue closedtensorflow/tfjs

Instanceof check fails in tf.Model.compile if two different versions are imported

To get help from the community, check out our Google group.

TensorFlow.js version

~0.12

Browser version

Chrome latest

Describe the problem or feature request

If I instantiate a model, and then pass it to a library that calls model.compile, passing in its own tf.Optimizer, the instanceof check in https://github.com/tensorflow/tfjs-layers/blob/de37e29133e2bc11defd3f5931e9e6afcacb7004/src/engine/training.ts#L745 fails as the library's tf.Optimizer is made from a different instantiation of tfjs. For some reason, even I'm using @tensorflow/tfjs: "~0.12" in both the library and the library consumer's package.jsons, I still get this error.

Code to reproduce the bug / link to feature request

closed time in 18 days

aman-tiwari

issue commenttensorflow/tfjs

Instanceof check fails in tf.Model.compile if two different versions are imported

This is fixed by @nsthorat.

aman-tiwari

comment created time in 18 days

delete branch tensorflow/tfjs

delete branch : fix-wasm

delete time in 20 days

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 491afe74dbe1485c3205adda485d93c4b2a98759

[WASM] Update starter projects to depend on latest core and wasm (#2706) BUG, DOC

view details

push time in 20 days

PR merged tensorflow/tfjs

Reviewers
[WASM] Update starter projects to depend on latest core and wasm cla: yes

Fixes https://github.com/tensorflow/tfjs/issues/2649

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+34 -20

0 comment

4 changed files

dsmilkov

pr closed time in 20 days

issue closedtensorflow/tfjs

[WASM] Some supported ops in wasm backend are not available

Hello, I have encountered some issues while using the tfjs-backend-wasm package, some infos are as follows:

TensorFlow.js version

  • @tensorflow/tfjs-core@1.4.0

  • @tensorflow/tfjs-backend-wasm@1.5.1-alpha5

Browser version

  • Google Chrome: 79.0.3945.79 (Official Build) (64-bit)

Describe the problem or feature request

Since 'Softmax' is not supported in wasm backend, I want to simulate a sample softmax function with ops 'Exp', 'Sum' and 'Div'. But when I try it with the guidance of webpack and test in chrome's inspect, the following error occured. tfjs-wasm-issue

Code to reproduce the bug / link to feature request

I used the code in the folder tfjs/tfjs-backend-wasm/starter/webpack, and I upgrade the package '@tensorflow/tfjs-backend-wasm' to version:1.5.1-alpha5 since I can't set the backend to 'wasm' while using the version in the 'yarn.lock'.

To reproduce the error, just change the code to the usage of other kernel like 'exp' or 'sum' like the followings: const x = tf.tensor1d([1, 2, 3]); x.sum().print(); // or tf.sum(x)

Then use commend yarn build & yarn watch to run the code in browser and the error occured. It seems like caused by the build or registry process or maybe I make some mistake in the usage. So I'm wondering are there any strategy to fix the issue?

Thanks.

closed time in 20 days

akineeic

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 073a4b7d9be519fd2ee8d5d749e29025804911da

save

view details

push time in 20 days

issue commenttensorflow/tfjs

[WASM] Some supported ops in wasm backend are not available

Hi, I just updated our starter projects to depend on latest wasm AND latest core in https://github.com/tensorflow/tfjs/pull/2706. The problem you were seeing is b/c tfjs-backend-wasm was at 1.5.1 but tfjs-core was at 1.4.0 (where 'exp' was not yet modularized) and that introduced a skew.

akineeic

comment created time in 20 days

PR opened tensorflow/tfjs

Reviewers
Update the starter project to depend on latest wasm

Fixes https://github.com/tensorflow/tfjs/issues/2649

+35 -20

0 comment

5 changed files

pr created time in 20 days

push eventtensorflow/tfjs

Daniel Smilkov

commit sha b3821ea2adf18a4477fac98df48d811342678989

save

view details

push time in 20 days

create barnchtensorflow/tfjs

branch : fix-wasm

created branch time in 20 days

push eventtensorflow/tfjs

Nikhil Thorat

commit sha 931e0c10306722c2e403cf8cc04cc13c0e68da04

[WASM] Update cdn link to point to the .js file. (#2690) BUG Fixes https://github.com/tensorflow/tfjs/issues/2683 Without this the CDN 404s because it tries to fetch the .wasm path relative to the base jsdelivr link (jsdelivr does not redirect). The alternative is we use tf.setWasmPath here.

view details

push time in a month

delete branch tensorflow/tfjs

delete branch : wasm-readme

delete time in a month

PR merged tensorflow/tfjs

[WASM] Update cdn link to point to the .js file. cla: yes

Fixes https://github.com/tensorflow/tfjs/issues/2683

Without this the CDN 404s because it tries to fetch the .wasm path relative to the base jsdelivr link (jsdelivr does not redirect).

The alternative is we use tf.setWasmPath here.

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+1 -1

0 comment

1 changed file

nsthorat

pr closed time in a month

issue closedtensorflow/tfjs

[WASM] `tfjs-backend-wasm` seems to request wasm file from the wrong location

I tried running the simple demo example in the tfjs-backend-wasm readme just now:

<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs"></script>
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-wasm"></script>

<script>
  (async function() {
    await tf.setBackend('wasm');
    ...
  })();
</script>

But I get a 404 for the wasm file that the script tries to request:

https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-wasm.wasm

I think it should instead be:

https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-wasm/dist/tfjs-backend-wasm.wasm

because that returns the wasm file, and this code does work:

...
await tf.wasm.setWasmPath("https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-wasm/dist/tfjs-backend-wasm.wasm");
await tf.setBackend('wasm');
...

closed time in a month

josephrocca

Pull request review commenttensorflow/tfjs

Update tfjs-react-native to 0.2.

+/** @license See the LICENSE file. */++// This code is auto-generated, do not modify this file!+const version = '0.2';+export {version};

re-export version from index.ts to be consistent with the other packages

tafsiri

comment created time in a month

issue commenttensorflow/tfjs

An error occured after using tf.slice function

Hi,

does this problem show up on Windows 10 as well? Also can you share info on what type of GPU/Intel CPU you have? Thanks!

dataCoderX10

comment created time in a month

issue closedtensorflow/tfjs

Using the video frames as a texture in tf.fromPixels without canvas element

TensorFlow.js version

master

Describe the feature request

I was looking for a way to feed video frames into tfjs without sampling from the video element manually through canvas. According to this MDN Tutorial Animating_textures_in_WebGL it should be possible by binding the HTMLVideoElement to the gl.texImage2D directly.

Here is a shortened snippet from the linked MDN page:

const video = document.createElement('video');
...
gl.texImage2D(gl.TEXTURE_2D, level, internalFormat, srcFormat, srcType, video)

I am not sure if this is implemented or if you have discussed this internally already. However, when looking at the tf.fromPixels GPU implementation in the current master branch it seems that you read the HTMLVideoElement data via canvas.

Would be really great to have this feature in tfjs ;)

Best, Christoph

closed time in a month

chaosmail

issue commenttensorflow/tfjs

Using the video frames as a texture in tf.fromPixels without canvas element

Sorry for the long delay on this one. We used to feed VideoElement directly to texImage2D but we had a problem that the actual pixel values are slightly different then when you read from a canvas.

This means that we get numerically different results when running webgl vs cpu vs wasm. For these reasons we decided to always feed into intermediate canvas which leads to consistent pixel values, at the cost of speed.

I'm going to close this issue since we don't have plans to address this in the near future, but feel free to comment with use-cases where you find fromPixels being a bottleneck, which will help revisit this in the future

chaosmail

comment created time in a month

MemberEvent
MemberEvent

delete branch tensorflow/tfjs

delete branch : update-main-doc

delete time in a month

push eventtensorflow/tfjs

Daniel Smilkov

commit sha cb7ed0788cdb6880d2afd1ed7d5ec9008a3c153c

Update main README.md with all the packages in the mono repo (#2604) DOC

view details

push time in a month

PR merged tensorflow/tfjs

Update main README.md with all the packages in the mono repo cla: yes

DOC

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+16 -2

0 comment

1 changed file

dsmilkov

pr closed time in a month

push eventtensorflow/tfjs

Xu Xing

commit sha 8f2d5ac47a1df58be8305ec5b85aed85e1281ab4

[webgpu] Fix memory leak in buffer manager (#2525) BUG * [webgpu] Fix memory leak in buffer manager * [webgpu] Fix memory leak in buffer manager * Fix typo Co-authored-by: Ann Yuan <annyuan@google.com>

view details

NALLEIN

commit sha b8caa142002bbf426e723a3ca4953a324a0e7115

Add reduce and made some changes as requested (#2617) FEATURE * Add reduce and made some changes as requested * Made some changes as requested * Made some changes to minimize imports from non-public API * Update backend_webgpu.ts * Update backend_util.ts * Update index.ts * Update backend_webgpu.ts Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Jiajia Qin

commit sha e1592d7313b4484b356356f77ee5ed859160473f

webgpu: Optimize conv2d (#2620) PERF * Swap x/filter sequence * Fix the crash when tileInner is smaller than workGroupSize.y * Optimize conv2d * Add comments Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Shanqing Cai

commit sha cf9c85d56e8ec7ae20a1c2d2d5f98f363774db87

Update benchmarks to tfjs 1.5.1 and python3 (#2630) * Update benchmarks to tfjs 1.5.1 and python3 DEV

view details

Nikhil Thorat

commit sha b9da49a35928dca75f41813e7606e6aacdd7a4cb

Use XNNPack for Sigmoid. (#2563) PERF The cache will always get hit because the key is simply 1. We keep the std::map so this code can be copy pasted without having to think.

view details

Kai Sasaki

commit sha 6a6e46d3121709181048aa2d145e25c840525788

Fix typo in profile doc (#2171) DOC

view details

dependabot[bot]

commit sha 89f75ee820d6ce2408346c1ba8047868a4ae6c0f

Bump handlebars from 4.1.2 to 4.5.3 in /tfjs (#2626) SECURITY Bumps [handlebars](https://github.com/wycats/handlebars.js) from 4.1.2 to 4.5.3. - [Release notes](https://github.com/wycats/handlebars.js/releases) - [Changelog](https://github.com/wycats/handlebars.js/blob/master/release-notes.md) - [Commits](https://github.com/wycats/handlebars.js/compare/v4.1.2...v4.5.3) Signed-off-by: dependabot[bot] <support@github.com>

view details

Ann Yuan

commit sha 88669b7dbfeedb60052be732fdb532f137cbed61

[webgpu] Link to tfjs-core@master (#2629) DEV

view details

Hao Yunfei

commit sha e1c60420e2a4f133028bd7d7aadcb7acae27772b

[webgpu] Simplify im2col to avoid using translate in matMul (#2621) PERF Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Na Li

commit sha b9a23bcdf6cf80a2a4c46bf27df0e4ffca7d3341

Add support to NonMaxSuppressionV5 (#2579) * Add NonMaxSuppressionV5 to tfjs-converter * Add support to NonMaxSuppressionV5 * Combine cases

view details

Nikhil Thorat

commit sha 09297d20dec2eef06150040a570499437757c919

[Core] Throw a better error for RMSPropOptimizer when no LR (#2634) DOC

view details

NALLEIN

commit sha 4a4529053df1d15472e54803c7869262231e4682

[webgpu] Add crop_and_resize kernel (#2636) FEATURE

view details

Na Li

commit sha 815ab9278c0ad65689c421ac52162c895c7dcc66

[WASM] Add NonMaxSuppressionV5 Op (#2598) FEATURE Add support to NonMaxSuppressionV5

view details

Na Li

commit sha 7ba9a7e11d84cfd8d295eb25b33e28ba22dcce92

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

view details

Jiajia Qin

commit sha 677d48e4ccc834dcce66c9f4435a62c156d129db

webgpu: Increase work per thread for maxpool (#2628) PERF With this change, maxPool[1, 131, 131, 64] has 50%~90% speedup on different platforms. * Add workPerThread to shaderKey Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Kangyi Zhang

commit sha c2d556415db8d9fb611a69a77a6288678ab0f86d

[node]Add node binding api to get number of loaded SavedModel (#2635) FEATURE * add binding api * update test * lint

view details

Ping Yu

commit sha 642a5419195de3747ed696757c6907f5ae188cd6

update the tensorflow deps to 2.1.0 (#2627) INTERNAL * update the tensorflow deps to 2.0.0 * fix lint * use 2.1.0rc2 due to an error on error saved model api * fix integration tests * fix pylint error and clean up the pip test * fix the pip package test * relex the pip requirement for h5py * update to 2.1.0

view details

Paul Van Eck

commit sha d0988eedf7bee90fb597e86f0231783e14f70565

Fix doc inconsistencies in tfjs-node image (#2658) DOC

view details

Yannick Assogba

commit sha 695f4aa6f8a6d259cc10e8901ca617dcaacad1dc

[tfjs-react-native] Add functions for streaming camera data into tensors. Adds functions for converting camera data to tensors for realtime use in model predictions. Primary goal of this change is to enable real time use of camera data with models like posenet/bodypix etc.

view details

Daniel Smilkov

commit sha 674ca1103bcf92a43b23b7488f670a56e2c680f8

Merge branch 'master' into update-main-doc

view details

push time in a month

Pull request review commenttensorflow/tfjs

Add reduce and made some changes as requested

 export {RMSPropOptimizer} from './optimizers/rmsprop_optimizer'; export {SGDOptimizer} from './optimizers/sgd_optimizer'; export {Scalar, Tensor, Tensor1D, Tensor2D, Tensor3D, Tensor4D, Tensor5D, TensorBuffer, Variable} from './tensor'; export {GradSaveFunc, NamedTensorMap, TensorContainer, TensorContainerArray, TensorContainerObject} from './tensor_types';-export {DataType, DataTypeMap, DataValues, Rank, RecursiveArray, ShapeMap, TensorLike} from './types';+export {DataType, DataTypeMap, DataValues, Rank, RecursiveArray, ShapeMap, sumOutType, TensorLike} from './types';

this introduces a top-level tf.sumOutType method, which we don't want as part of our API since we are aligning with TF Python. All the other exports from types are just interfaces/types that do not exist at runtime. We should instead expose sumOutType as part of the backend_util namespace.

NALLEIN

comment created time in a month

delete branch tensorflow/tfjs

delete branch : rms-lr

delete time in a month

create barnchtensorflow/tfjs

branch : rms-lr

created branch time in a month

push eventhaoyunfeix/tfjs

Ping Yu

commit sha 642a5419195de3747ed696757c6907f5ae188cd6

update the tensorflow deps to 2.1.0 (#2627) INTERNAL * update the tensorflow deps to 2.0.0 * fix lint * use 2.1.0rc2 due to an error on error saved model api * fix integration tests * fix pylint error and clean up the pip test * fix the pip package test * relex the pip requirement for h5py * update to 2.1.0

view details

Paul Van Eck

commit sha d0988eedf7bee90fb597e86f0231783e14f70565

Fix doc inconsistencies in tfjs-node image (#2658) DOC

view details

Daniel Smilkov

commit sha 4394c8b5c1688bd42f73ea68cfea3ab0a038d277

Merge branch 'master' into fix_2654

view details

push time in a month

push eventWenheLI/tfjs

NALLEIN

commit sha 4a4529053df1d15472e54803c7869262231e4682

[webgpu] Add crop_and_resize kernel (#2636) FEATURE

view details

Na Li

commit sha 815ab9278c0ad65689c421ac52162c895c7dcc66

[WASM] Add NonMaxSuppressionV5 Op (#2598) FEATURE Add support to NonMaxSuppressionV5

view details

Na Li

commit sha 7ba9a7e11d84cfd8d295eb25b33e28ba22dcce92

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

view details

Jiajia Qin

commit sha 677d48e4ccc834dcce66c9f4435a62c156d129db

webgpu: Increase work per thread for maxpool (#2628) PERF With this change, maxPool[1, 131, 131, 64] has 50%~90% speedup on different platforms. * Add workPerThread to shaderKey Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Daniel Smilkov

commit sha 2b13645bffa3cb757a0889ada9f84704da6d87c1

Merge branch 'master' into fix-save-db

view details

push time in a month

issue commenttensorflow/tfjs

Add kernel for image standardization.

Good point. In that case, custom op on a per-model basis SGTM (we can't add it to official API until TF Python has it)

annxingyuan

comment created time in a month

issue openedgoogle/pytype

Add support for custom typings of third_party dependencies

I have a third-party dependency on apache-beam and I'd like to add minimal pyi typings locally, only for the small API that I'm using.

I noticed that pyright has a "typings" config to specify a directory with custom typings. Is there an analogous option in pytype?

created time in a month

startedgoogle/pytype

started time in a month

pull request commenttensorflow/tfjs

[WASM] Add NonMaxSuppressionV5 Op

Thanks for adding a comment on top of that line! Nice work Na!

lina128

comment created time in a month

pull request commenttensorflow/tfjs

[WASM] Add NonMaxSuppressionV5 Op

Yes we do need another free since backend.wasm._free(resOffset) will only remove the 4 bytes allocated for the pointer address of pSelectedScores, but it will not remove the array of values that were allocated on the heap that the pointer points to, thus you need explicit removal of that. The reason why you do not need to do the same with pSelectedIndices is because we wrap that in a Tensor object, which gets globally tracked and later gets disposed by the tidy mechanisms when it gets out of scope.

lina128

comment created time in 2 months

push eventtensorflow/tfjs

Nikhil Thorat

commit sha 9a44bc1ef4d21912d039928b15d821ad8b4df934

[bench] Update benchmarks to show aggregate kernel timing or in… (#2595) DEV

view details

Xu Xing

commit sha c5eaf89764b886e842c16027e590dfcf1e57635d

[webgpu] Add stridedSlice kernel (#2441) FEATURE Fix: https://github.com/tensorflow/tfjs/issues/2363 Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Na Li

commit sha 0bddd3c7f2836c95bc246b02ec8fa33d92849569

Add NonMaxSuppressionV5 to tfjs-node (#2549) Add the op using modular pattern Create new folder for all new kernels Note: Tested with latest core at e5a0c84

view details

Na Li

commit sha f6a7635d22eb867abce6b6e78256517ff1f25e7e

Update tfjs-core to 1.5.1. (#2605)

view details

Na Li

commit sha 333406582c9353bd1df65118845fc35ade007cd5

Update tfjs-layers, tfjs-converter to 1.5.1, 1.5.1. (#2606) MISC

view details

Kangyi Zhang

commit sha 1bf4a698c3038e4e515e8cd4617b4612595ffbcd

Update tfjs-data to 1.5.1. (#2607)

view details

Na Li

commit sha 045b53623bee3a289d36af9b188b0c660618a6e8

Update tfjs to 1.5.1. (#2608)

view details

Kangyi Zhang

commit sha 6febe5dcb3d40cef226d3770505ffb96c8241e34

Update tfjs-node, tfjs-node-gpu to 1.5.1, 1.5.1. (#2609) INTERNAL

view details

Xu Xing

commit sha 3fb35297ef4572e444d257341e4bccdb0a3f866a

[webgpu] Add fusedConv2d kernel (#2448) FEATURE Fix: https://github.com/tensorflow/tfjs/issues/2362 Co-authored-by: Ann Yuan <annyuan@google.com>

view details

Kangyi Zhang

commit sha 8bddb6fd243f9d99cffb2792e6279e3e84e127e8

Update tfjs-backend-wasm to 1.5.1-alpha5. (#2610) INTERNAL * Update tfjs-backend-wasm to 1.5.1-alpha1. * update alpha5 * save Co-authored-by: Na Li <linazhao128@gmail.com>

view details

Ann Yuan

commit sha 44c13bad73a49fae1f0f961c89506f4d24bdbd26

[wasm] Add ScatterND kernel. (#2600) FEATURE

view details

Ann Yuan

commit sha 1fa1beb3b08e00065b830bc2dcec435bbf5a8af3

[wasm] Add Gather, GatherNd kernels. (#2601) FEATURE

view details

Daniel Smilkov

commit sha da4bdf9caa7f56416529a8415353668d402d4606

Merge branch 'master' into update-main-doc

view details

push time in 2 months

PR closed tensorflow/tfjs-models

DO NOT SUBMIT Debug CocoSSD with WASM cla: yes

<!-- Reviewable:start --> This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+800 -673

0 comment

8 changed files

dsmilkov

pr closed time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha a3201cc64688097a37e0b70d9fcee6e2be708209

save

view details

push time in 2 months

create barnchtensorflow/tfjs

branch : codegen

created branch time in 2 months

delete branch tensorflow/tfjs

delete branch : codegen

delete time in 2 months

create barnchtensorflow/tfjs

branch : codegen

created branch time in 2 months

push eventtensorflow/tfjs

Nikhil Thorat

commit sha cdd1db58c569f8a21ad3ee303e5168c82c3295a3

[WASM] Update FAQ with link to github issues for feedback. (#2602) DOC Also update setBackend documentation with WASM info (this wont happen until the next release).

view details

Daniel Smilkov

commit sha d2c4d7dfe09d001b32d7fa246cbe7a4079992119

Merge branch 'master' into update-main-doc

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 816d37667584c8ff74d613ea1e336341fdb8715c

save

view details

Daniel Smilkov

commit sha 669fd4c6ef626af204d12f9a0f8373913137af14

save

view details

push time in 2 months

PR opened tensorflow/tfjs

Reviewers
Update main README.md with all the packages in the mono repo

DOC

+16 -2

0 comment

1 changed file

pr created time in 2 months

create barnchtensorflow/tfjs

branch : update-main-doc

created branch time in 2 months

delete branch tensorflow/tfjs

delete branch : update-wasm-doc

delete time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha ccfe6de2bb97d4593bff571bbe22f67d1c20748a

[wasm] Mention which models are support by the wasm backend (#2603) DOC

view details

push time in 2 months

PR merged tensorflow/tfjs

[wasm] Mention which models are support by the wasm backend cla: yes

<!-- Reviewable:start --> This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+21 -27

0 comment

3 changed files

dsmilkov

pr closed time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha aaa8acef28faebfb09d36fcca1ff8add92f4b0f7

save

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 6247833dc9b654eb343691359677e8ea0b3f4c0c

save

view details

push time in 2 months

PR opened tensorflow/tfjs

Reviewers
[wasm] Mention which models are support by the wasm backend
+17 -23

0 comment

3 changed files

pr created time in 2 months

create barnchtensorflow/tfjs

branch : update-wasm-doc

created branch time in 2 months

delete branch tensorflow/tfjs

delete branch : grad_registry

delete time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha b71f6bffbdd2e1037d14042a950aa28757881477

[core] Add gradient registry (#2584) Add API to register gradients in a global registry, analogous to how we register modular kernels. FEATURE

view details

push time in 2 months

PR merged tensorflow/tfjs

[core] Add gradient registry cla: yes

Add API to register gradients in a global registry, analogous to how we register modular kernels.

FEATURE

<!-- Reviewable:start -->

This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+271 -69

1 comment

10 changed files

dsmilkov

pr closed time in 2 months

pull request commenttensorflow/tfjs

[core] Add gradient registry

@pyu10055 PTAL when you get a chance! Thanks!

dsmilkov

comment created time in 2 months

pull request commenttensorflow/tfjs-models

Add blazeface model.

This is in great shape. Really polished! Great work

annxingyuan

comment created time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 4d6feb125035fb0c1b5c2819d540dc1b01164ad7

[wasm] Update benchmark section. Clarify that CPU is plain JS (#2585) DOC

view details

Nikhil Thorat

commit sha fbafe5d037246f8a01ccf7acec9c26b940e69fb6

[WASM] Add Sin, Cos, Tanh, Rsqrt. (#2589) FEATURE

view details

Daniel Smilkov

commit sha b705dae9dc09cb5ced7f19f8049cd4a05f39905b

Merge branch 'master' into grad_registry

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 1187b1696de671a93402f5bd30dd313c44eddf63

save

view details

push time in 2 months

delete branch tensorflow/tfjs

delete branch : update-wasm-bunch

delete time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 4d6feb125035fb0c1b5c2819d540dc1b01164ad7

[wasm] Update benchmark section. Clarify that CPU is plain JS (#2585) DOC

view details

push time in 2 months

PR merged tensorflow/tfjs

[wasm] Update benchmark section. Clarify that CPU is plain JS cla: yes

Preview: https://github.com/tensorflow/tfjs/blob/update-wasm-bunch/tfjs-backend-wasm/README.md

<!-- Reviewable:start --> This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+29 -28

0 comment

4 changed files

dsmilkov

pr closed time in 2 months

PR opened tensorflow/tfjs-models

DO NOT SUBMIT Debug CocoSSD with WASM
+800 -673

0 comment

8 changed files

pr created time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha f3bbf0884de2601db918d675736ef2442bc8c946

save

view details

push time in 2 months

push eventtensorflow/tfjs-models

Daniel Smilkov

commit sha ed9690c7523cf4b6e8c527ad515789f7dcf4c294

save

view details

push time in 2 months

create barnchtensorflow/tfjs-models

branch : fix-wasm-ssd

created branch time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 1cdd6fae999850d2cf00a7ec9e8f9ccc73dd5e6b

save

view details

push time in 2 months

PR opened tensorflow/tfjs

Reviewers
[wasm] Update benchmark section. Clarify that CPU is plain JS
+22 -23

0 comment

3 changed files

pr created time in 2 months

create barnchtensorflow/tfjs

branch : update-wasm-bunch

created branch time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 4950ae5fb5485dfe93c30425fe5568e0060ff46f

save

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 1117fe0014aab31f7eab0a60125eedf513c27198

save

view details

push time in 2 months

PR opened tensorflow/tfjs

WIP Add gradient registry

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

+194 -110

0 comment

6 changed files

pr created time in 2 months

create barnchtensorflow/tfjs

branch : grad_registry

created branch time in 2 months

delete branch tensorflow/tfjs

delete branch : improve-bench

delete time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha b97dbea6b217d5bc957ba40bbc34f5ff60a8ae84

[core] Dispose all webgl programs when disposing the backend. (#2577) BUG

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 39f73a39bb3a66e331046c611d45a14c62300300

save

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 17bd87ffa15944ea693c32af0c3720e56b10d88c

save

view details

push time in 2 months

PR opened tensorflow/tfjs

Reviewers
[core] Dispose all webgl programs when disposing the backend.
+78 -3

0 comment

2 changed files

pr created time in 2 months

create barnchtensorflow/tfjs

branch : improve-bench

created branch time in 2 months

delete branch tensorflow/tfjs

delete branch : wasm-update-bench

delete time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha 7053fb0b99948b9ff9e0c9a1501fa5f9ed3cb11d

[wasm] Update benchmarks with Pixel 3 and Macbook Pro 15 numbers (#2573) DOC

view details

push time in 2 months

PR merged tensorflow/tfjs

Reviewers
[wasm] Update benchmarks with Pixel 3 and Macbook Pro 15 numbers cla: yes

Preview: https://github.com/tensorflow/tfjs/blob/wasm-update-bench/tfjs-backend-wasm/README.md

<!-- Reviewable:start --> This change is <img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/> <!-- Reviewable:end -->

+18 -15

0 comment

4 changed files

dsmilkov

pr closed time in 2 months

push eventtensorflow/tfjs

Nikhil Thorat

commit sha 4bab4649461343904cc1ce9a68b162406db530aa

[WASM] Add Sum. (#2566) FEATURE Most of the implementation is taken from Max.

view details

翠 / green

commit sha 7f5676781215a754fe21cffc01e623a455348a0b

[WASM] Add Greater, GreaterEqual kernels (#2543) FEATURE

view details

Daniel Smilkov

commit sha cf20422f02c4bf8a422564f26fed2b857e36066b

Merge branch 'master' into wasm-update-bench

view details

push time in 2 months

push eventtensorflow/tfjs

Daniel Smilkov

commit sha fd0e81361748c6b733edb764c94e5686ff5e4808

save

view details

push time in 2 months

more