search

facebookresearch / DiT

Official PyTorch Implementation of "Scalable Diffusion Models with Transformers"
Other
6.31k stars 564 forks source link

Prompt-conditioning model instead of class-conditioning #70

Open anarabiyev opened 8 months ago

anarabiyev commented 8 months ago

"We provide a training script for DiT in train.py. This script can be used to train class-conditional DiT models, but it can be easily modified to support other types of conditioning."

Does anyone have a clue about modifying the model to the prompt-conditioning model?

wuzelei123 commented 8 months ago

Can you use self-trained data to generate images?

anarabiyev commented 8 months ago

Can you use self-trained data to generate images?

I didn't understand what you meant. My purpose is to modify this model so that I can feed prompt as input instead of class labels.

wuzelei123 commented 8 months ago

The error occurred when I used the weight file generated during the official code training in sample.py. Like： RuntimeError: CUDA error: device-side assert triggered Compile with TORCH_USE_CUDA_DSA to enable device-side assertions. What parameters did you modify during training?

anarabiyev commented 8 months ago

I haven't modified any parameters yet, I am trying to analyze code and find a clue where to start.

wuzelei123 commented 8 months ago

okok，thank you. I've just solved my problem.

twinkleyang1 commented 6 days ago

@wuzelei123 Hello, I have the same problem, how did you solve it。The error occurred when I used the weight file generated during the official code training in sample.py. Like： RuntimeError: CUDA error: device-side assert triggered Compile with to enable device-side assertions. What parameters did you modify during training?TORCH_USE_CUDA_DSA