diff --git a/server/text_generation_server/models/__init__.py b/server/text_generation_server/models/__init__.py index 85c98bfd..f8150b5e 100644 --- a/server/text_generation_server/models/__init__.py +++ b/server/text_generation_server/models/__init__.py @@ -1362,7 +1362,8 @@ def get_model( revision=revision, quantize=quantize, speculator=speculator, - dtype=torch.bfloat16, + dtype=dtype, + default_dtype=torch.bfloat16, kv_cache_dtype=kv_cache_dtype, trust_remote_code=trust_remote_code, lora_adapter_ids=lora_adapter_ids,