About the error running the train_retriever.py

[XiangyanChen]

Hi! I'm interested in your work.
After I ran encode_comments.py, retrieval.py, and recent.py from preprocess, I tried to run train_retriever.py, but I encountered a problem "RuntimeError: "clamp_min_cpu" not implemented for 'Half'" when running "style_reps = torch.nn.functional.normalize(style_reps, dim=-1)". I searched on Google, and it seems that the train_retriever.py was run on the CPU.
Since I'm not familiar with Pytorch, I seek help from you.

[shuailiu6626]

Hi! Thank you for your interest in our work! I'm really sorry for the late response. We use the Trainer class from the transformers library (version 4.17.0) in train_retriever.py which should automatically move the model to CUDA if CUDA is available. Could you please check whether you have CUDA and PyTorch set up correctly by running python -c "import torch; print(torch.cuda.is_available())" from command line?