thu-ml / SageAttention

Quantized Attention that achieves speedups of 2.1x and 2.7x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.
BSD 3-Clause "New" or "Revised" License
400 stars 17 forks source link

Please create a ComfyUI node to use SageAttention #37

Open wardensc2 opened 1 week ago

wardensc2 commented 1 week ago

Hi @jt-zhang

Thank you so much for your work however for normal people like us it's very difficult to understand your code and implement it on normal you. Can you write a Comfy Node for us to use Sage Attention easier

Thank in advance

jason-huang03 commented 6 days ago

Sorry we are not familiar with Comfy. However, I have read blogs that claim to use SageAttention with Comfy like this. Perhaps you can refer to them. The Comfy community is quite active and I believe there are ready solutions to the problem.

jason-huang03 commented 6 days ago

Also the mochi node supports using SageAttention.