kvablack ddpo-pytorch issues

kvablack / ddpo-pytorch

DDPO for finetuning diffusion models, implemented in PyTorch with LoRA support

MIT License

388 stars 41 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

training-code

#25 ParnyanAtaei opened 2 weeks ago
0
OOM when using "stabilityai/stable-diffusion-2-1" with batch size of 2

#24 xilongzhou opened 4 months ago
1
Questions about the reward curve and bert.

#23 zjuAIHz opened 5 months ago
1
Finetuning on google colab

#22 alirezanobakht13 opened 5 months ago
0
Closes #20 Reward Batch Size

#21 mao-code opened 5 months ago
0
Batch size unrecogonized

#20 mao-code opened 5 months ago
1
Code logics, thanks

#19 junyongyou opened 6 months ago
0
unet keeps producing nan during training

#18 EYcab opened 7 months ago
2
SDXL Support?

#17 rdcoder33 opened 11 months ago
1
Update README.md to include a note about the `trl` integration

#16 sayakpaul closed 11 months ago
0
prompt-dependent value function optimization

#15 hkunzhe opened 11 months ago
0
Support for other schedulers

#14 desaixie closed 11 months ago
1
Hello, when I trained an aesthetic model using the default configuration on 8 A800 cards, I found that the training process got stuck after completing one epoch, but it worked fine when using a single A800 card. May I ask what could be the cause of this situation?

#13 cjt222 closed 11 months ago
8
Question about the optimized objective.

#12 JacobYuan7 closed 12 months ago
1
About the training with prompt_image_alignment configuration which uses llava_bertscore reward function

#11 QZJ-2003 closed 11 months ago
5
Suggest to use larger gradient accumulation steps instead of multi GPUs

#10 hkunzhe closed 1 year ago
3
Only log rewards from process 0

#9 desaixie closed 1 year ago
1
fp16 only if using lora?

#8 GiilDe closed 1 year ago
2
About the large dataset and Unet Training

#7 liuyuemaicha closed 11 months ago
3
Prompt Alignment with LLaVA-server: Client-side prompt and image doesn't match server side reward

#6 desaixie closed 1 year ago
4
On reproducing LLaVA alignment experiments.

#5 bhattg closed 1 year ago
2
OOM despite using A100-80GB GPU and 128GB CPU memory (+16 CPUs per task)

#4 noshaba closed 1 year ago
2
reproducing the aesthetic experiment

#3 seashell123 closed 1 year ago
7
On reproducibility and LoRA

#2 bhattg closed 1 year ago
1
Gif visualization

#1 SnowdenLee closed 1 year ago
1