Is training FLUX with NF4 possible? #798
Replies: 6 comments 1 reply
-
we rely in third party libraries to do the quantisation because of how complex this gets while training. you will have to ask optimum-quanto devs that. also it already trains on consumer GPUs. |
Beta Was this translation helpful? Give feedback.
-
Hi, thank you for your reply. I would like to ask if it is possible to use |
Beta Was this translation helpful? Give feedback.
-
everything else the trainer depends on has to support it too. these are core dependencies like Accelerate and PEFT and the Diffusers projects. |
Beta Was this translation helpful? Give feedback.
-
I roughly understand what you mean. |
Beta Was this translation helpful? Give feedback.
-
yes that is inference related. trust me if it worked here sayak would open a pull request |
Beta Was this translation helpful? Give feedback.
-
@sayakpaul is indeed working on this :) |
Beta Was this translation helpful? Give feedback.
-
I saw on Reddit that someone achieved significant memory savings and speed improvements using the NF4 version of FLUX on Forge.
![image](https://private-user-images.githubusercontent.com/79357052/356943492-c22331fc-88b6-4078-a1a5-8614e0b406cf.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.h7QIeB9DyX9Ss8Mvf671fTvc2R9V7Ghks2OTrpFkOBo)
The quality impact wasn't very significant.
I wanted to ask if introducing NF4 in training is feasible.
Perhaps it could enable FLUX training on consumer-grade GPUs.
Beta Was this translation helpful? Give feedback.
All reactions