Glossary: Vision and Image Generation

dtype

Also written: torch_dtype

The setting for what number format a model's weights are stored in, such as `torch.float32` or `torch.bfloat16`.