Only enable fp16 on z image models that actually support it. (#12065)
This commit is contained in:
@@ -1093,7 +1093,7 @@ class ZImage(Lumina2):
|
||||
|
||||
def __init__(self, unet_config):
|
||||
super().__init__(unet_config)
|
||||
if comfy.model_management.extended_fp16_support():
|
||||
if comfy.model_management.extended_fp16_support() and unet_config.get("allow_fp16", False):
|
||||
self.supported_inference_dtypes = self.supported_inference_dtypes.copy()
|
||||
self.supported_inference_dtypes.insert(1, torch.float16)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user