search

Mikubill / naifu

Train generative models with pytorch lightning
MIT License
294 stars 38 forks source link

DPO training issues #32

Closed DonStroganotti closed 7 months ago

DonStroganotti commented 7 months ago

I wanted to give DPO training a try but there seems to be multiple issues with the PairedDataset.

reso and interp are not assigned to self in init but self.reso is being used later in the file

https://github.com/Mikubill/naifu/blob/af5c34b338e121787b8fa4299456129d7e310a92/data/paired_wds.py#L54

Getting this error when using the default train_dpo.yaml config:


  File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop
    data = fetcher.fetch(index)
  File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/usr/local/lib/python3.10/dist-packages/torch/utils/data/_utils/fetch.py", line 51, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/workspace/naifu/data/paired_wds.py", line 63, in __getitem__
    example = self.preprocess_train(example)
  File "/workspace/naifu/data/paired_wds.py", line 33, in preprocess_train
    images = [
  File "/workspace/naifu/data/paired_wds.py", line 34, in <listcomp>
    Image.open(io.BytesIO(im_bytes)).convert("RGB")
TypeError: a bytes-like object is required, not 'int'```
Mikubill commented 7 months ago

Fixed in bf1db75, thanks for the issue!