Multimodal — CUDA / Python report

Every llama-cpp-python horror story starts with the same question: which wheel do I actually need? The answer depends on three things you've probably never printed together - your Python version, your PyTorch CUDA version, and whether the llama-cpp-python you have installed exports Qwen3VLChatHandler. This node is a one-button answer to all three.

It's the smallest possible utility: one boolean input (run, default true) and one output (report, a STRING). Toggle run, and the node probes your environment and writes a report you can read on the node itself or wire into a Show Text node. It reports the python.exe path (so you can see whether you're installing into ComfyUI's environment or some other Python that ComfyUI will never look at), maps torch.version.cuda to the cuXXX tag naming convention (12.4 → cu124), names your GPU, and - the part that saves real time - checks whether Qwen3VLChatHandler is importable in your current install.

Why that last bit matters: the parent pack's flagship node uses GGUF + mmproj vision through llama-cpp-python, and Qwen3-VL vision specifically needs a build that exports Qwen3VLChatHandler. Stock pip wheels often don't have it, and the symptom - a model that loads fine but can't see images or produces empty replies - is easy to misdiagnose as a weights problem. This node turns "is it my model or my wheel?" into a yes/no you can read in seconds.

The workflow is: run this report first, copy the cuXXX hint, then hand it to the pack's "Multimodal - llama-cpp wheel pick (GitHub)" node (or the python -m llm_comfy_multimodal.install_llama_wheel CLI), which finds a wheel whose filename matches your CPython tag, platform, and CUDA tag. If the report already shows Qwen3VLChatHandler: OK, you're done - no wheel shopping needed, the GGUF path is ready.

It's a diagnostic, not a workhorse, so don't expect to build workflows around it. You'll reach for it twice: once during setup, and again after any Python/CUDA upgrade that mysteriously breaks vision. For a node whose entire job is to print facts about your machine, it's remarkably good at preventing the specific facepalm that this pack's dependency chain is set up to cause.