Testing on Mac (cpu-only)

[bojan2110]

First, thanks for sharing this valuable project! We are very excited to test these models in our research at Erasmus Rotterdam.

My question is related to running the models on cpu-only mode, as I have a mac without nvidia (so no cuda support). I am trying to reproduce the sentiment analysis project (dbrd) using the finetune_dbrd.ipynb notebook you have shared. However, when training the model with the provided code I get a message (at the end of the training process) AssertionError: Torch not compiled with CUDA enabled.

So my question is if there is a way to run this with cpu-only pytorch? I see that in your train.py script you are recognising the device type (args.device) and I have tested that it prints out 'cpu'. Are there some parameters that I need to change in order to run the notebook on my mac laptop?

[iPieter]

First of all, thanks for your interest in RobBERT!

The answer to your question depends on why you want to use those notebooks:

• To actually replicate our results. If so, I would really recommend installing the old (2019) dependencies on a machine with a 1080 Ti. Otherwise, I don’t think you’ll be able to get the same results. There exist a lot of examples of small differences between GPUs and CPUs (and especially an ARM M1) to illustrate that this is not an ideal setting to replicate our results. Also, I do not have an M-series mac, so I cannot test this myself...
• Finetuning of your own model. For this, that notebook is also hopelessly outdated. Huggingface now has a very nice Trainer api, I would recommend using that. If I now finetune a model, I also use the Trainer in combination with a training logger (I use MLFlow with some custom hyperparameter sweeps, but weights and biases or comet.ml are other options). This tutorial should set you on the right path.

Good luck!

[bojan2110]

Thanks a lot, in that case I will skip the replication and focus on finetuning the RobBERT model for our needs via Huggingface.