Freepik

/

flux.1-lite-8B-alpha

like 402

Follow

freepik 150

are there any plans to release a 12B version?

are there any plans to unrelease this version?

do we need all the blocks cant we compress the blocks into fewer blocks while predicting the same weights etc, iam sure theres some technic for it

Are there any plans to release a 4B version?

We are working on making available for 24Gb VRAM GPUs, without modifying weights precision.

We are working on making available for 24Gb VRAM GPUs, without modifying weights precision.

do try more prompt variety and maybe using other data rather than flux outputs , it can collapse training a model on its own outputs

also a 4-8 step destil lora would be cool too