Sylvain Gugger

Yes, that's a current limitation of the `dispatch_batches=True` option. I will try to dive into this and find a better solution for batches of different sizes, but that's going to...

Could you explain the context to me? There is no need to install TensorFlow on colab (that might actually mess up things more than help), and if the user is...

The environment in which the notebook is executed should be managed outside of the notebook by the user themselves. I think this should be replaced by a markdown cell pointing...

We don't have support for SLURM right now :-)

It should all work as long as you can launch it!

Yes, that's exactly what Accelerate does for both sharded and non-sharded models. Not sure what feature you feel is missing, could you share some code?

@Chris-hughes10 it was automatically closed by the stale bot due to the absence of activity. @NouamaneTazi is still investigating whether this could have some negative impact sometimes.

We could indeed support `device_placement=True` or a list of strings (like `["model", "optimizer"]`) to allow for different placements.

Sorry, it took me a bit of time but the PR above introduces the API you requested!

Hi there! It's probably not compatible with fastai as the fastai library uses their own `DataLoader` and `Optimizer` classes (different from the PyTorch ones). We're thinking of how we can...