LoRA Stat Viewer (Yogurt Nodes)

Every serious tool in this pack's LoRA suite asks you for a regex. key_pattern, .* by default - and if you leave it at that, you're telling the node "operate on literally every key." That's fine for a global operation, but the moment you want to compress only the attention layers, or scale only one block, you need to know what your LoRA's keys actually look like. YogurtLoraStatViewer is the x-ray: it dumps the key structure so you can write the regex with your eyes open.

It's part of ComfyUI-YogurtNodes, yogurt7771's 150+ node all-in-one pack - obscure, MIT-licensed, auto-generated README - and it's the diagnostic half of the LoRA surgery suite, the one you run before touching the scalers and compressors.

How it works

One required input, lora - the pack's LORA object, typically from YogurtLoadLoraOnly. Two options:

• show_all_keys - a boolean, default off. When on, it appends every key name to the output. That can be a wall of text on a big LoRA, so leave it off unless you really need the full list.
• sample_count - an INT, default 12, up to 200. How many keys to show as a sample.

Output is a single lora_info STRING: the key patterns, counts, and structure you need to design a filter.

What you're actually looking for

The big question this answers is what kind of LoRA you have. The pack's rank operations (merge, compress) work on standard .lora_down / .lora_up pairs. DoRA, LoCon, and reshape variants have different key layouts - lora_a/lora_b vs lora_down/lora_up, extra conv keys, magnitude tensors for DoRA. One look at the sample keys tells you whether a given LoRA is eligible for compression or whether it'll get skipped.

It's also how you learn the block naming so a regex like .*\.attn.* or .*to_out.* actually matches what you expect. Run this first, write the regex second, and you'll stop guessing.

Where you'll use it

• Before YogurtLoraRankCompress, to confirm the LoRA is standard-pair and to pick a layer filter.
• Before YogurtLoraScaleWeights / YogurtLoraScaleAlpha, to build a key_pattern that targets one module.
• Just to satisfy curiosity about what a downloaded LoRA actually contains.

Install

ComfyUI Manager, search ComfyUI-YogurtNodes, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes
cd ComfyUI-YogurtNodes
pip install -r requirements.txt

Restart, look under "Yogurt Nodes". No model downloads - it reads the LoRA you hand it. Light dependencies, and the only "cost" is the patience to read a key dump instead of guessing.