Even the 8B model makes many consumer-grade graphics cards unable to run smoothly

#3
by jian2023 - opened

Perhaps 6B or 4B, or perhaps the fp8 version of 8B will make flux easier to use

Hi @jian2023 , this is the alpha version and the plan is to soon release the 24Gb compatible model, with even less number of parameters

@dverdu-freepik Fits perfectly fine for 24gb vram on comfyui, if clip and vae are forced to cpu. Runs fine even if I don’t do that, I just like the extra room for Lora’s and upscaling.

The amount of parameters is perfect, so I’m hoping this amount, 8b will be kept and improved while smaller and larger parameter sizes are also offered. As far as I’m concerned this is the 24gb compatible model. However, we always welcome file size decrease, speed increase, with minimum to no quality loss and any future improvements.

Maybe something can be done to make the clip sizes smaller while maintaining 1:1 quality.

fp8 pweeez

Hi @jian2023 , this is the alpha version and the plan is to soon release the 24Gb compatible model, with even less number of parameters
I only have 16G video memory, can not load directly, will the FP8 version? Or wait for a smaller GGUF?

We need this model but in FP8 imo.

Sign up or log in to comment