[Web] Error using opus-mt-mul fp16 models with WebGPU

[mram0509]

Describe the issue

An unknown error occurs when loading "https://huggingface.co/Xenova/opus-mt-mul-en/resolve/main/onnx/encoder_model_fp16.onnx" and "https://huggingface.co/Xenova/opus-mt-mul-en/resolve/main/onnx/decoder_model_merged_fp16.onnx" models. The opus-mt-mul-en fp32 models load correctly and there are no errors.

To reproduce

Create a translate inference pipeline using "@huggingface/transformers" version 3.3.3, which uses onnxruntime-web - version 1.21.0-dev.20250206-d981b153d3. The error occurs when loading the - https://huggingface.co/Xenova/opus-mt-mul-en/resolve/main/onnx/encoder_model_fp16.onnx

Urgency

No response

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.21.0-dev.20250206-d981b153d3

Execution Provider

'webgpu' (WebGPU)