PyTorch Version Checker

You updated something - or followed a Colab guide, or installed ComfyUI into your own Python instead of the portable build - and now you genuinely don't know which PyTorch is loaded. That's the exact situation this node exists for. It's the diagnostic half of the ComfyUI-PyTorch210Patcher pack: it reads your actual torch and CUDA versions and tells you whether the pack's companion patcher applies.

What it does

The PyTorch Version Checker is gloriously boring. It grabs torch.__version__ and torch.version.cuda, compares the version against 2.10, prints a neat banner to the console, and returns three values:

• pytorch_version - a string like "2.10.0"
• cuda_version - a string like "12.6" (or "N/A", more below)
• needs_patch - a boolean, True when your torch is 2.10 or newer

It has no inputs at all and it's an output node, so you just drop it on the canvas, run, and read. It doesn't change a thing - deliberately. The whole job is to be the answer key for its sibling.

Why you'd reach for it

Mainly as step one of any "my Wan workflow just broke" debug. PyTorch 2.10 is real - it's the ROCm 7-era release people were chasing for big performance gains on AMD - and it's also the one that breaks WanVideoModel in Kijai's ComfyUI-WanVideoWrapper. When needs_patch comes back True, that's your cue to drop the PyTorch 2.10 Compatibility Patcher in front of the WanVideo model loader and move on with your life.

The quieter value: it confirms which Python environment ComfyUI is really using. If the version here disagrees with what you think you installed, trust the node, not your memory. That alone saves the classic hour of reinstalling torch into the wrong venv.

Where people get burned

• cuda_version says "N/A" while you clearly have a GPU. That means your torch build was compiled without CUDA support - a bigger problem than this pack fixes, and usually a reinstall-your-torch situation.
• needs_patch only means the WanVideoModel break. It says nothing about whether your install is otherwise healthy. Don't read a True as "everything is broken," and don't read a False as "everything is fine."

Installing it

It ships in the ComfyUI-PyTorch210Patcher pack, so install is the same as its sibling: search "ComfyUI-PyTorch210Patcher" in ComfyUI Manager and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/huyl3-cpu/ComfyUI-PyTorch210Patcher.git

Then restart ComfyUI. No requirements.txt, no model downloads, nothing heavy - it's a couple hundred lines of plain Python. You only need Kijai's ComfyUI-WanVideoWrapper installed if you actually plan to run the patcher half of the pack.

Honest verdict: it's a utility node in the best sense - does one boring thing well, then you take it out of the workflow. Keep it installed as a debug tool, not as a permanent resident of your graph.