DarkAlchy

yeah, I just found that out. Kohya is 12.4 cuda and 2.4 torch but comfyui is 12.1 and 2.3.0

> yeah if you want --fast you need pytorch 2.4 or later on windows. I recommend pytorch nightly, you can grab a standalone with pytorch nightly package here if you...

I lost a few nodes in comfy that demand FA2 (even if not used) so I had to roll back to 2.1.2 just so I can use the nodes. Florence2...

I have Cuda 12.6 and 12.8 and the same for either (Linux and Windows). Not sure how they ended up releasing this like this, so I figured it was something...

Damn, as I have the fp8 model. Guess this is now a dead project until we get a gguf version.

full or fp8 same nonsense? :( I spent a day in hell down a deep rabbit hole to fix a few nodes. I got everything to work BUT this thing....

> This is because when you use fp8 version, you cannot skip the q8_kernel node, otherwise you will get this, also make sure you are on cuda 12.8 when you...

I read somewhere where ComfyAnon has decided to make NF4 deprecated in favour of GGUF. I use NF4 myself, but haven't in the last few days (was nice and way...

On Windows if you are in the directory with the export_trt.py the required way to path is "../../../" not the single "..\" fractalfantasy said because, for me at least, I...

13B being a horrible bug riddled release, I hope so, but I doubt it. Seems a lot of us can't even run this new one. Not sure if because fp8...