nom

Results 17 comments of nom

> i have compel SD3 90% of the way there.. Any update?

@mattt Yes for me it still reproduces with latest build

Any progress on this @yiyixuxu @sayakpaul ? Still feels like a huge gap between Comfy and diffusers, especially with new samplers coming out like the now very popular [RES4LYF](https://github.com/ClownsharkBatwing/RES4LYF/) (e.g....

I suspect people here will want code and weights, not a demo

Somehow ``` loss = loss / accelerator.gradient_accumulation_steps accelerator.deepspeed_engine_wrapped.engine.backward(loss) ``` does not give equivalent results to `accelerator.backward(loss)` with deepspeed. What gives?