Have a ComfyUI workflow? Get a readable Workflow Doc in your browser.Make a workflow readable.Try it now ↗
A ComfyUI Ksampler modification to capture latents per step. These can then be used to measure the quality of various quantization methods.
Package profile
README
Run the evaluation after collecting your latents with the EvalSampler via:
python MeasureResults.py --eval-dir /Comfy/output/eval_latents --stratify-std
It expects the baseline latents to be in a sub-folder called "BF16". It then compares against every other subfolder. Results will be printed in terminal, and saved as various formats.
Nodes in this pack
3 nodesSources
3 sourcesVerified Sep 13
Source excerpts
1 excerptSource context: Repo BobJohnson24/ComfyUI-EvalSampler
Verified Aug 7
Verified Sep 13
