OOM Error
#80
by feel-123 - opened
I'm encountering an Out-of-Memory (OOM) error while running the FLUX.1-Kontext-dev model. I understand this model has high GPU memory requirements.
Could you please suggest an alternative version or a lighter variant of the FLUX.1-Kontext-dev model that offers comparable performance but can run efficiently on GPUs with ≤16 GB of VRAM, without significant loss in accuracy?
You can try the INT4 quantized version（just search for it in the model tree). I previously managed to run it successfully on a 24GB GPU without enabling offload, so you might want to try using it with offload enabled.