QwenImage Pipeline

[saddam213]

It looks like there will soon be support for Qwen2_5_VL models in onnx so I should be able to add support for QwenImage models (hopefully)

Qwen2_5_VL model support: microsoft/onnxruntime-genai#1882

Model Conversions

• - Pipelines
• - AutoEncoder (AutoencoderKLQwenImage)
• - TextEncoder (Qwen2_5_VLForConditionalGeneration)
• - Transformer (QwenImageTransformer2DModel)

https://huggingface.co/TensorStack/QwenImage-onnx

[saddam213]

AutoEncoder conversion fails with::

torch.onnx.errors.UnsupportedOperatorError: Exporting the operator 'aten::_upsample_nearest_exact2d' to ONNX opset version 20 is not supported

Transformer conversion fails with:

torch.onnx.errors.UnsupportedOperatorError: Exporting the operator 'aten::view_as_complex' to ONNX opset version 20 is not supported

[saddam213]

vae_decoder converted @ float32

CPU - pass
CUDA - pass
DirectML - fail