Add Qwen3-VL export support and example runner

[seyeong-han]

Companion PR - Optimum-Executorch - Add Qwen3-VL export support for multimodal text-to-text pipeline

Overview

Adds export and runtime support for Qwen3-VL-2B-Instruct, a 2.2B parameter vision-language model. Export goes through optimum-executorch via the existing multimodal-text-to-text task, producing a single .pte with vision_encoder, text_decoder, and token_embedding methods.

The optimum-executorch changes (in a companion PR) handle three Qwen3-VL-specific concerns during torch.export: pre-computing M-RoPE vision positions that use data-dependent ops, injecting position_ids via a forward hook so the text decoder export doesn't hit get_rope_index, and falling back to AutoModelForImageTextToText when AutoModelForPreTraining doesn't resolve.

This PR adds the ExecuTorch-side example:

• examples/models/qwen3_vl/run_qwen3_vl.py — Python runtime that loads the .pte via ExecuTorchModule.run_method, driving token_embedding and text_decoder through ExecuTorch. The vision encoder runs in PyTorch eager because the portable runtime's aten::convolution.out does not yet support 5D inputs (Conv3d).
• examples/models/qwen3_vl/README.md — Export command, runtime usage, method shapes, quantization config, and architecture notes.

Quantized model is ~1.4 GB (8da4w decoder, 8da4w/8da8w encoder, 8w embeddings).
Decode rate is ~25 tokens/sec on Apple Silicon M-series via XNNPACK.

Run Qwen3-VL-2B

python qwen3/run_qwen3_vl.py \
  --model_path qwen3/Qwen3-VL-2B-Instruct-xnnpack/model.pte \
  --image_path qwen3/test_image.jpg \
  --prompt "What is in this image?" \
  --max_new_tokens 200

Output

Image embeddings shape: torch.Size([651, 2048])
Prefill done: 667 tokens in 4.12s

Prompt: What is in this image?
--------------------------------------------------
Response: This image shows a wooden dock or pier extending into a calm, blue lake. The dock is constructed from weathered wooden planks and leads from the foreground toward the center of the lake. On the right side of the dock, there is a metal railing, and on the left side, there is a metal ladder. The lake is surrounded by a dense forest of green trees, and in the background, there are snow-capped mountains under a cloudy sky. The entire scene is captured in a wide, horizontal view, with the dock as the main focal point.
--------------------------------------------------
Prompt tokens:    666
Generated tokens: 114
Prefill time:     4.119s
Decode rate:      26.59 tokens/sec
Total time:       8.369s

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17572

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

⚠️ 8 Awaiting Approval

As of commit bac415e with merge base a24d3e7 ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

• Build documentation (gh)
• Build Presets (gh)
• Lint (gh)
• periodic (gh)
• pull (gh)
• Test CUDA Builds (gh)
• Test CUDA Windows Export (gh)
• Test Metal Backend (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.