Open zhousu220 opened 10 months ago
Check run.sh to see the whole data generation process. You need first use convert_train.py to generate file the same size as your gpu num (1/2/4/8). Then you need to generate validation set (dev set) use build_valid.py. Training.py only use dev-0.npy for quick validation. Validate.py will use all dev-{}.npy (same size to gpu num).
According to the idea you provided, after carefully reading the code, I finally succeeded in running through the code. Thank you very much for your help.
When training the model, the dev-0.npy file is not generated, and the file in the data/raw directory displayed in the error message is searched, as shown in the figure, only the train-x.npy file is displayed. Maybe it's because the model training requires four GPUs.