Q8 would be cool
especially since the vaes are out now, i could probably squeeze more frames out of my 5090.
today ,will upload it
ensure that you keep the vaes out of it before creating the Q8.
ensure that you keep the vaes out of it before creating the Q8.
https://huggingface.co/Kijai/LTXV2_comfy here it is seperated
ensure that you keep the vaes out of it before creating the Q8.
https://huggingface.co/Kijai/LTXV2_comfy here it is seperated
the quality for a Q8 model is better than fp8.
ensure that you keep the vaes out of it before creating the Q8.
https://huggingface.co/Kijai/LTXV2_comfy here it is seperated
the quality for a Q8 model is better than fp8.
Yes it might be true but I am just guiding him so we can have a proper GGUF. He should use the transformer only model for making a gguf instead of the checkpoints in the ltx repo
vae and audio_vae will be f32,it caused little big size but less error...