# Runtime tested in the pytorch/pytorch:2.4.1-cuda11.8 image (see README).
torch==2.4.1
numpy
transformers==4.44.0   # only needed for the GPT-2-scale experiments (gpt2_jlens.py)