Sven Killig

I have 96 GB RAM, and indeed ComfyUI fills it up when inferencing flux2.
b.t.w., there's obvious spam in the issue tracker since 2 months.

I already did, but wasn't sure, because then I got

torch.AcceleratorError: CUDA error: an illegal memory access was encountered
Search for `cudaErrorIllegalAddress' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.

I wonder what ComfyUI uses. It inferences in about a minute - even with fp8 and a local text encoder.

I hoped enable_group_offload would speed it up a bit more than

>python3 flux.2-with-remote-text-encoder.py
torch.__version__='2.9.1+cu130'
diffusers.__version__='0.36.0.dev0'
Using GPU: NVIDIA GeForce RTX 5070 Ti
Total VRAM: 15 GBs
Running remote text encoder ☁️
Done ✅
  2%|███▏  | 1/50 [02:29<2:02:25, 149.92s/it]

It's just a copy of https://github.com/huggingface/blog/blob/main/flux-2.md?plain=1#L283 but here you go: https://gist.github.com/sonic74/423c03483fbc13e7fd99ac97bcec8ff8

No, it's an undefined variable in line 30 of the enable_group_offload example.

https://github.com/huggingface/blog/blob/main/flux-2.md?plain=1#L283 causes

    transformer_id, subfolder="transformer", torch_dtype=torch_dtype, device_map="cpu"
    ^^^^^^^^^^^^^^
NameError: name 'transformer_id' is not defined