tmoroder

Fine tuning TensorFlow DeBERTa fails on TPU

9

### System Info Latest version of transformers, Colab TPU, tensorflow 2. - Colab TPU - transformers: 4.21.0 - tensorflow: 2.8.2 / 2.6.2 - Python 3.7 ### Who can help? @LysandreJik,...

Wondering if the role must be assistant rather than user in the ``format_dpo_sample`` function of the training script ``run_dpo.py`` in ``chosen`` and ``rejected`` messages, e.g., [line 124](https://github.com/philschmid/deep-learning-pytorch-huggingface/blob/59b37973074de90004d10e5ff636f98160c9743a/training/scripts/dpo/run_dpo.py#L124)?