We have been using TensorFlow for the past 5 years, but recently we have found this framework to be very heavy and slow. So, we started considering moving to PyTorch because it has a reputation for being easier to teach and understand than TensorFlow. Another reason is that we want to follow the work of LLaMa, our LLM, and they are all developed on PyTorch.

PyTorch is a framework that is rapidly gaining popularity in the deep learning community. Compared to other frameworks such as TensorFlow, PyTorch has a slower learning curve and is easier to adopt. This makes it a smooth switch for existing TensorFlow users to migrate to PyTorch.

Migrating to PyTorch is an important step in our quest to improve ease of use and performance in our deep learning projects. From an education and research perspective, PyTorch is an attractive option for us.