Nodes/Krea-2-SVDQuant-ComfyUI/Krea2 SVDQuant Diagnostics
ComfyUI Node

Krea2 SVDQuant Diagnostics

Passthrough node that reports which comfy_kitchen backend the quantized layers dispatch to, plus memory accounting and timings. Paste the output into bug reports.

By alperktt·Created about a month ago·Updated 5 days ago· 11
Krea2 SVDQuant Diagnostics
  • model
  • model
  • report
mode
tokens4096
CategoryKrea2/SVDQuant

Inputs (3)

NameTypeDefaultDescription
modelMODELOutput of the Krea2 SVDQuant W4A4 Loader.
modeCOMBOdispatch: which kernel actually runs (start here). env: versions and memory accounting. bench: quantized vs bf16 per layer. profile: torch.profiler table. compile: graph breaks a TorchCompileModel run would pay.
tokensINT409664–65536Sequence length to probe with. 4096 = 1024x1024. Kernel selection is shape-dependent, so match your real run.

Outputs (2)

NameTypeDescription
modelMODELThe model, unchanged - wire it on to your sampler.
reportSTRINGThe report as text, for pasting into a bug report.