onnxruntime: [Web/js/webgpu] `error: cannot assign to value expression of type 'u32'`

Describe the issue

I modified the transformers.js demo site to test the webgpu backend and it just logged many warnings and errors. I’ve removed duplicates and sent them below:

Tint WGSL reader failure: :69:14 error: cannot assign to value expression of type 'u32'
      indices[2] -= sizeInConcatAxis[inputIndex - 1u];
             ^

:69:7 note: 'let' variables are immutable
      indices[2] -= sizeInConcatAxis[inputIndex - 1u];
      ^^^^^^^

:65:9 note: let 'indices' declared here
    let indices = o2i_output(global_idx);
        ^^^^^^^


 - While validating [ShaderModuleDescriptor]
 - While calling [Device].CreateShaderModule([ShaderModuleDescriptor]).

[Invalid ShaderModule] is invalid.
 - While validating compute stage ([Invalid ShaderModule], entryPoint: main).
 - While calling [Device].CreateComputePipeline([ComputePipelineDescriptor]).

[Invalid ComputePipeline] is invalid.
 - While Validating GetBindGroupLayout (0) on [Invalid ComputePipeline]

Tint WGSL reader failure: :74:14 error: cannot assign to value expression of type 'u32'
      indices[1] -= sizeInConcatAxis[inputIndex - 1u];
             ^

:74:7 note: 'let' variables are immutable
      indices[1] -= sizeInConcatAxis[inputIndex - 1u];
      ^^^^^^^

:70:9 note: let 'indices' declared here
    let indices = o2i_output(global_idx);
        ^^^^^^^


 - While validating [ShaderModuleDescriptor]
 - While calling [Device].CreateShaderModule([ShaderModuleDescriptor]).

An uncaught WebGPU validation error was raised: [Invalid BindGroupLayout] is invalid.
 - While validating [BindGroupDescriptor] against [Invalid BindGroupLayout]
 - While calling [Device].CreateBindGroup([BindGroupDescriptor]).

An uncaught WebGPU validation error was raised: [Invalid CommandBuffer] is invalid.
 - While calling [Queue].Submit([[Invalid CommandBuffer]])

The output is also incorrect: “Hello, how are you?” -> “Bonjour, comment :”, should be “Bonjour, comment êtes-vous?” (which actually is quite close, so, not bad!)

To reproduce

1. Using yesterday’s CI build: https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=1104462&view=artifacts&pathAsName=false&type=publishedArtifacts

2. You can replace the import in the demo site (source code available here), but it should be thrown when running any example app.

3. Models tested: t5-small and whisper.tiny.en

Urgency

blocks webgpu release

ONNX Runtime Installation

Other / Unknown

ONNX Runtime Version or Commit ID

https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=1104462&view=artifacts&pathAsName=false&type=publishedArtifacts

Execution Provider

‘webgpu’ (WebGPU)