[qwen] Add VideoChatFlashQwenModel text-decoder builder

[Copilot]

Imports the Python text-decoder builder additions from onnxruntime-genai PR #2147 (file hash 11d14636…cfcb50 → src/python/py/models/builder.py) so mbext can export the text component of VideoChatFlashQwenForCausalLM (OpenGVLab/VideoChat-Flash, Qwen2.5-7B backbone).

Changes

• modelbuilder/builders/qwen.py — new VideoChatFlashQwenModel(QwenModel):

  • Sets self.model_type = "VideoChat_Flash_QwenForCausalLM" so the base class derives the genai_config type videochat_flash_qwen.
  • Overrides load_weights to call Qwen2ForCausalLM.from_pretrained directly, bypassing the upstream VCF AutoConfig which pulls in av/cv2/decord/imageio.

• modelbuilder/builder.py — dispatch case for VideoChatFlashQwenForCausalLM that forces extra_options["exclude_embeds"] = True (decoder consumes inputs_embeds from the embedding-merger that fuses InternVideo2 visual tokens).

• tests/fast/test_random_videochat_flash_qwen.py — fast offline test (fp32/fp16/int4 CPU + fp16 CUDA) following the test_random_gemma3_conditional.py pattern:

  • Builds a tiny random Qwen2 model, rewrites architectures in the saved config.json to VideoChatFlashQwenForCausalLM (since Qwen2ForCausalLM.save_pretrained would otherwise overwrite it) to exercise the new dispatch.
  • Asserts the exported ONNX exposes inputs_embeds (not input_ids) and runs the prefill/decode discrepancy check with embed_fn=model.model.embed_tokens.

The VLM vision encoder and embedding-merger remain out of scope, matching upstream Phase 1.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• https://api.github.com/repos/microsoft/onnxruntime-genai/pulls/2147
  • Triggering command: /usr/bin/curl curl -sL -H Accept: application/vnd.github.v3.diff REDACTED -o /tmp/pr2147.diff (http block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[github-actions]

📂 Impacted Files

3 file(s) changed in this pull request:

🔵 modelbuilder/builder.py (+6 / -0)
🔵 modelbuilder/builders/qwen.py (+32 / -0)
🟢 tests/fast/test_random_videochat_flash_qwen.py (+150 / -0)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.86%. Comparing base (d8cd68a) to head (e7e226a).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #371      +/-   ##
==========================================
+ Coverage   78.82%   78.86%   +0.03%     
==========================================
  Files          27       27              
  Lines        8058     8071      +13     
==========================================
+ Hits         6352     6365      +13     
  Misses       1706     1706              

Flag	Coverage Δ
fast-tests	78.86% <100.00%> (+0.03%)	⬆️
fast-tests-ort-nightly	78.81% <100.00%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.