First off, thank you for your hard work on this package! It's proven to be _very_ helpful for fully leveraging multi-GPU training of large-scale language models.
Is your feature request related to a problem? Please describe.
I'm using the (also incredible) transformers package by the folks at HuggingFace to do a number of tasks, including text classification. I realized that training these models can be quite computationally expensive, even on decently-sized single-GPU instances. I found out about this repo and was (with relative ease) able to train the same model over 4 GPUs and 16-bit precision much, _much_ more easily than training with vanilla PyTorch.
I was wondering if there was any interest in a contributed tutorial that specifically goes through the process of integrating pytorch-lightning with transformers, which seems to have quite a bit of staying power and success within the realm of applied NLP. Regardless, thanks again!
Describe the solution you'd like
N/A?
Describe alternatives you've considered
N/A?
Additional context
N/A?
yes! that would be amazing. would welcome a PR!
Neat! I noticed that the current domain template (for GANs here) is a Python script — is that preferred over a full notebook?
@williamFalcon bumping this — I've got a more in-depth notebook that shows how to use transformers with pytorch-lightning, but I can make it a Python script, instead. Let me know!
I would suggest creating a tutorial folder with notebooks. Later we can automatically generate documentation from these notebooks, see example from readthedocs
@dataframing go ahead and make a notebook! That would be awesome
Sounds good! Apologies for the delay here; had a busy week. Should be able to get something out soon 👍
@dataframing any updates? in the meantime, we can use the tutorial i put together on medium and the COLAB in the readme
Most helpful comment
I would suggest creating a tutorial folder with notebooks. Later we can automatically generate documentation from these notebooks, see example from readthedocs