Reproducibility

[hassancpu]

Hi there,

I use the library to fine-tune some models for our task. However, I realized that I get different results when I repeat the training with the same hyperparameters(lr, batch size, seed, ...) and without any augmentation. I also check the inputs in each iteration, which are the same. I would appreciate any suggestion in this regard. BTW, I can share the log files if you think they could be useful.