Yang Gu comments

Results 29 comments of


                                            Yang Gu

fromPixels has different results on CPU and GPU backends

There are 2 problems to be solved so that we can fix the correctness tests in e2e: 1) willReadFrequently is set to true for GPU backends (WebGL and WebGPU), so...

LayersModel#predict() results in all zeros when using WebGPU backend in Deno

> Oh, seems function must declare before entry point **AND** before `@compute @workgroup_size(256, 1, 1)` Shader as below is workable. > > ```js > ... > var sharedBuf : array;...

major difference in performance between sync and async methods to download tensor data

@mattsoulanille Thanks and let's wait for Emscripten's feedback at https://github.com/emscripten-core/emscripten/issues/17844.

major difference in performance between sync and async methods to download tensor data

Sorry for the late response. MessageChannel also works and I updated my test (https://github.com/webatintel/webatintel.github.io/blob/main/timer.html#L44) to double confirm this. Given @vladmandic already fixed the issue in Emscripten upstream (thank you so...

Customize setTimeout

@pyu10055 @mattsoulanille, given related Emscripten issues were fixed (Thanks @vladmandic !) and @mattsoulanille upgraded Emscripten to include the fixes, can we merge this change now?

Remove fromPixelsAsync

@mattsoulanille Can we add this into next major release (5.0)?

TensorFlowJS fails reliably on Intel GPU running Windows, otherwise succeeds

This looks like an Intel specific issue, so we'll investigate this.

An error occurred during model execution: "RangeError: offset is out of bounds".

I just checked I can run this model correctly with either wasm or webgpu backend.

[Web] The YOLOv8 segmentation model with batching option is not runing on the GPU ?

Really sorry for the late reply! Currently in WebGPU EP implementation, there is a limitation that axis in softmax has to the last. We will soon fix this.