Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
The torch_dtype argument can be
used to initialize the model in half-precision on a CUDA device only.