Radial Attention loads extremely slowly, unlike Sage Attention which loads much faster.

[peepeepeepoopoopoo]

I have an RTX 4060 8GB computer, 16GB RAM, updated NVIDIA & ComfyUI, free hard drive space, and an RTX 3070 8GB laptop, 24GB RAM. I've tested with just ComfyUI open, right out of the box. Both have the correct installation process as indicated: https://github.com/woct0rdho/ComfyUI-RadialAttn.

When I tested with Sage Attention, it ran a thousand times faster. I've attached the workflow I'm using with RADIAL: https://transfer.it/t/RpsYsQhFQBiD

CMD:

[ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
FETCH DATA from: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json [DONE]
[ComfyUI-Manager] All startup tasks have been completed.
C:\ComfyUI\comfy\samplers.py:955: UserWarning: PYTORCH_CUDA_ALLOC_CONF is deprecated, use PYTORCH_ALLOC_CONF instead (Triggered internally at C:\actions-runner_work\pytorch\pytorch\pytorch\c10\core\AllocatorConfig.cpp:28.)
if latent_image is not None and torch.count_nonzero(latent_image) > 0: #Don't shift the empty latent image.
(RES4LYF) rk_type: res_2s
25%|██████████████████████████████████████████████████ | 1/4 [07:30<22:31, 450.47s/it]

[woct0rdho]

See woct0rdho/ComfyUI-RadialAttn#7 . Before testing RadialAttention, it's identified that there is an issue with torch.compile and ComfyUI LoRA in PyTorch >= 2.8 .