diff --git a/server/text_generation_server/models/__init__.py b/server/text_generation_server/models/__init__.py index 5b1b5715..ab3b25b7 100644 --- a/server/text_generation_server/models/__init__.py +++ b/server/text_generation_server/models/__init__.py @@ -87,7 +87,9 @@ def get_model( trust_remote_code: bool, ) -> Model: if dtype is None: - dtype = torch.float16 + # Keep it as default for now and let + # every model resolve their own default dtype. + dtype = None elif dtype == "float16": dtype = torch.float16 elif dtype == "bfloat16":