Can this FP8 model be deployed on 4090? How is the speed?
#1
by
yoolv
- opened
Can this FP8 model be deployed on 4090? How is the speed?
Yes, I made this fork with some changes to juggle LLM/VAE/DiT, it should consume 13-17 GB depending on image resolution: https://github.com/rkfg/Step1X-Edit
Expect 20-30s for 512 size and 1.5 minutes for 1024. I can only estimate as I have a 3090 Ti, I get 50s for 512 and 2.5 min for 1024.
请问这个模型的官方工作流在哪里