benzstation

Results 4 comments of benzstation

manual cast: torch.bfloat16 when using fp8 combined flux.dev models causing vram issues with LoRAs

Got same issue - its causing horrible generation speed (20s/it instead of 2s/it) on a regular RTX 4070 12GB

manual cast: torch.bfloat16 when using fp8 combined flux.dev models causing vram issues with LoRAs

> There have been several updates and rollbacks over the past few days. Have you updated to the latest version and is the same issue still occurring? Yup: ### ComfyUI...

manual cast: torch.bfloat16 when using fp8 combined flux.dev models causing vram issues with LoRAs

> There have been several updates and rollbacks over the past few days. Have you updated to the latest version and is the same issue still occurring? For what its...

manual cast: torch.bfloat16 when using fp8 combined flux.dev models causing vram issues with LoRAs

Tested other mix of versions: Total VRAM 12282 MB, total RAM 32691 MB pytorch version: **_2.3.1+cu121_** Set vram state to: LOW_VRAM Device: cuda:0 NVIDIA GeForce RTX 4070 : cudaMallocAsync Using...