Software Engineering Courses (SECourses)•16mo ago

@Dr. Furkan Gözükara any update on the fp8 for dreambooth Furak Obama?

MaxiviOP•9/21/24, 6:31 PM

@Dr. Furkan Gözükara do you get this issue when extracting lora

flux_extract_lora.py:68
31%|█████████████████████████████████████████████████████████████████████████████ | 93/304 [01:57<09:26, 1.86s/it]
Traceback (most recent call last):
File "/workspace/kohya_ss/sd-scripts/networks/flux_extract_lora.py", line 219, in <module>
svd(**vars(args))
File "/workspace/kohya_ss/sd-scripts/networks/flux_extract_lora.py", line 87, in svd
value_t = f_tuned.get_tensor(key)
File "/workspace/kohya_ss/sd-scripts/library/utils.py", line 249, in get_tensor
return self._deserialize_tensor(tensor_bytes, metadata)
File "/workspace/kohya_ss/sd-scripts/library/utils.py", line 265, in _deserialize_tensor
byte_tensor = torch.frombuffer(tensor_bytes, dtype=torch.uint8)
ValueError: both buffer length (0) and count (-1) must not be 0