I'm calling it, PyTorch is no longer the go-to framework for deep learning research. Transformers and JAX have finally caught up in terms of innovation and ease of use, and I'm not convinced the PyTorch community is adapting quickly enough to the shift.