shliu0

about dataloader through prepare()

6

In [tutorial](https://huggingface.co/docs/accelerate/quicktour), it is mentioned that `Some data at the end of the dataset may be duplicated so the batch can be divided equally among all workers.` So if my...

[Q] why some steps didn't be logged during training?

2

 I am training a vqvae and using 1*8 gpus, the code to log as belows, where variable `logs` is a dict: ``` if accelerator.sync_gradients and not...

question about ema_vocab_hit_SV

1

In quant.py, I saw that you use ema_vocab_hit_SV rather than vocab_hit_V to record codebook hit frequency, what is the reason for doing this?