pydantic
# transformers is not needed for the model code, but is used for:
# - tokenizers
# - parity-testing against HF implementations
# - loading HF model weights
transformers
einops