- Create docs/VLM_SUPPORT_NOTES.md with research findings for future
VLM (Vision-Language Model) support using SDNQ-quantized models
- Remove Qwen3-VL-32B-Instruct from model catalog as it's a VLM that
outputs text (not images) and requires transformers, not diffusers
- Add comment in registry explaining why VLM is excluded
VLM models like Qwen3-VL require a separate SDNQCaptioner node
implementation using the transformers library. The notes document
the loading pattern, inference code, and proposed node design.
Claude
8cdca188f6