Added data generation & training support for qwen3 vl model

[shanjiaz]

vl model support:

• Extract text_config from multimodal model configs in vocab mapping, training, and model initialization
• Added kwargs to support vision language parameters like deepstack_input_embeds in custom worker for Qwen3-VL's multimodal processing
• Use vLLM's native get_language_model() method to access the text model in vision-language architectures
• Detect base model structure for vision-language models

Util changes:

• Add suffix pattern matching for weight keys (e.g., "embed_tokens.weight" matches "model.embed_tokens.weight")

• Support single-file safetensors models (no index.json) by building virtual weight maps

• Run e2e test for a qwen 3 vl model

[github-actions]

📦 Build Artifacts Available
The build artifacts (`.whl` and `.tar.gz`) have been successfully generated and are available for download: https://github.com/vllm-project/speculators/actions/runs/20759905084/artifacts/5041230401.
They will be retained for up to 30 days.
Commit: 9c5031d

[github-actions]

Summary

Status	Count
🔍 Total	158
✅ Successful	155
⏳ Timeouts	0
🔀 Redirected	2
👻 Excluded	0
❓ Unknown	0
🚫 Errors	1
⛔ Unsupported	0

Errors per input

Errors in docs/developer/code-of-conduct.md

• [ERROR] https://www.contributor-covenant.org/translations/ | Network error: Connection reset by server. Server forcibly closed connection

Redirects per input

Redirects in CONTRIBUTING.md

• [200] https://docs.pytest.org/ | Redirect: Followed 1 redirect resolving to the final status of: OK. Redirects: https://docs.pytest.org/ --> https://docs.pytest.org/en/stable/

Redirects in docs/developer/contributing.md

• [200] https://docs.pytest.org/ | Redirect: Followed 1 redirect resolving to the final status of: OK. Redirects: https://docs.pytest.org/ --> https://docs.pytest.org/en/stable/
  Full Github Actions output

[fynnsu]

Looks good!

[rahul-tuli]

LGTM! great job