Comfyui?
#1
by deleted - opened
Hello! Thank you for your interest in our work!
DisCa is quite different from traditional LoRA. It primarily introduces a small model (a lightweight cache predictor as mentioned in paper) that performs alternate inference with the large main model, thereby achieving further inference acceleration on top of distilled models.
As this is currently an exploration of a new academic approach, it cannot be directly applied to ComfyUI at the moment. However, as this research direction matures in the future, there may be opportunities to integrate it.