tritant

Results 7 comments of tritant

FLUX - NF4 uses a lot of RAM and VRAM when a LORA is being used.

Same here, dev nf4 v2 + lora = generation time x10 (4060 ti 16 gb)

FLUX - NF4 uses a lot of RAM and VRAM when a LORA is being used.

Without lora time is 50 secs, with a fp8 version, it works well, not tested with gguf ![Capture d'écran 2024-08-17 141109](https://github.com/user-attachments/assets/90412af9-0831-4042-888a-1e01cdfd2480) When finished ![Capture d'écran 2024-08-17 144544](https://github.com/user-attachments/assets/c0c9bd1b-4804-4c7f-9ab1-b818bcc8f55e)

FLUX - NF4 uses a lot of RAM and VRAM when a LORA is being used.

With gguf it working well ![Capture d'écran 2024-08-17 150500](https://github.com/user-attachments/assets/ec24056a-094a-4026-a06c-be681bc8ea6a)

Controlnet for Flux

And to use flux fill with inpainting, is there a solution?

problem with torch and cuda

RTX 5000 compatibility install cu128 https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/2608

Bug: Flux Distilled CFG Scale sticks after first generation regardless of subsequent changes

Same here

Bug: Flux Distilled CFG Scale sticks after first generation regardless of subsequent changes

No fix for that? it's really very restrictive.