Please give us a fp8 Text encoder(non diffusers format)

#5
by kakkkarotto - opened

Please give us a fp8 Text encoder(non diffusers format)

I've converted it and uploaded it to huggingface https://huggingface.co/GitMylo/LTX-2-comfy_gemma_fp8_e4m3fn/tree/main

Could it be downscaled further? I can run the model just fine using the bigger clip and it only encodes using cpu but cant get these to work on 12gb vram.

nvm --novram forced cpu and it is faster at this size.

Sign up or log in to comment