NotImplementedError when clip.load

[DtYXs]

When running model, preprocess = clip.load("ViT-B/32", device=device)

D:\Anaconda\anaconda3\envs\A\lib\site-packages\torch\serialization.py:591: UserWarning: 'torch.load' received a zip file that looks like a TorchScript archive dispatching to 'torch.jit.load' (call 'torch.jit.load' directly to silence this warning)
" silence this warning)", UserWarning)
Traceback (most recent call last):
File "D:\python\helpers\pydev\pydevd.py", line 1483, in _exec
pydev_imports.execfile(file, globals, locals) # execute the script
File "D:\python\helpers\pydev_pydev_imps_pydev_execfile.py", line 18, in execfile
exec(compile(contents+"\n", file, 'exec'), glob, loc)
File "D:\models.py", line 191, in
model = CLIP(cfg)
File "D:\models.py", line 173, in init
self.model, self.preprocess = clip.load("ViT-B/32", device=device)
File "D:\Anaconda\anaconda3\envs\A\lib\site-packages\clip\clip.py", line 135, in load
model = build_model(state_dict or model.state_dict()).to(device)
File "D:\Anaconda\anaconda3\envs\A\lib\site-packages\clip\model.py", line 396, in build_model
vit = "visual.proj" in state_dict
File "D:\Anaconda\anaconda3\envs\A\lib\site-packages\torch\jit_script.py", line 624, in contains
return self.forward_magic_method("contains", key)
File "D:\Anaconda\anaconda3\envs\A\lib\site-packages\torch\jit_script.py", line 611, in forward_magic_method
raise NotImplementedError()
NotImplementedError

How do I solve this problem?

[haofanwang]

Try model, preprocess = clip.load("ViT-B/32", device=device, jit=False)

[DtYXs]

Try model, preprocess = clip.load("ViT-B/32", device=device, jit=False)

The problem still occurs. Beacuse jit defaults to Fasle.

It seems that the state_dict in vit = "visual.proj" in state_dict is torch.jit._script.RecursiveScriptModule class,
and when call "visual.proj" in state_dict, the error occurred.

My environment and version are executed according to the commands in the document. But it is in Windows. Does it matter?

[haofanwang]

CLIP/clip/clip.py

Line 125 in 573315e

model = torch.jit.load(model_path, map_location=device if jit else "cpu").eval()

It seems that this line fails, a runtime exception is raised. Could you move it before try, do you still meet the same issue?

[DtYXs]

CLIP/clip/clip.py

Line 125 in 573315e

model = torch.jit.load(model_path, map_location=device if jit else "cpu").eval()

It seems that this line fails, a runtime exception is raised. Could you move it before try, do you still meet the same issue?

Thank you very much for your reply. I move it, and

Traceback (most recent call last):
File "D:\python\helpers\pydev\pydevd.py", line 1483, in _exec
pydev_imports.execfile(file, globals, locals) # execute the script
File "D:\python\helpers\pydev_pydev_imps_pydev_execfile.py", line 18, in execfile
exec(compile(contents+"\n", file, 'exec'), glob, loc)
File "D:/models.py", line 172, in init
self.model, self.preprocess = clip.load("ViT-B/32", device=self.cfg.device, jit=False)
File "D:\Anaconda\anaconda3\envs\A\lib\site-packages\clip\clip.py", line 122, in load
model = torch.jit.load(model_path, map_location=device if jit else "cpu").eval()
File "D:\Anaconda\anaconda3\envs\A\lib\site-packages\torch\jit_serialization.py", line 161, in load
cpp_module = torch._C.import_ir_module(cu, str(f), map_location, _extra_files)
RuntimeError: open file failed because of errno 42 on fopen: Illegal byte sequence, file path: C:\Users\Administrator/.cache/clip\ViT-B-32.pt
python-BaseException

I think the problem comes from my CUDA version is 11.1.
Since pytorch seems to have no cudatookit for 11.1 (I tried conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.1 but PackagesNotFoundError: The following packages are not available from current channels: cudatoolkit=11.1). Then I tried conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0 and conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.3 two environments. Each environment has problems.

[ZZZ9116]

Hi, I have got the same problem. Did you solve it?

[DtYXs]

Hi, I have got the same problem. Did you solve it?

Actually not, I still cannot run it in my Windows system Cuda11.1 Pytorch1.7.1.
But I can run it in the other server Linux Ubuntu18.04.5 Cuda11.1 Python3.7.9 Pytorch1.9.1.

[rohaldb]

I am also having this issue - any update?

[liuxingbaoyu]

Maybe because the windows username contains non-ascii characters, this is a pytorch bug

Modify clip.py

CLIP/clip/clip.py

Lines 125 to 134 in 40f5484

	try:
	# loading JIT archive
	model = torch.jit.load(model_path, map_location=device if jit else "cpu").eval()
	state_dict = None
	except RuntimeError:
	# loading saved state dict
	if jit:
	warnings.warn(f"File {model_path} is not a JIT archive. Loading as a state dict instead")
	jit = False
	state_dict = torch.load(model_path, map_location="cpu")

    with open(model_path, 'rb') as opened_file:
        try:
            # loading JIT archive
            model = torch.jit.load(opened_file, map_location=device if jit else "cpu").eval()
            state_dict = None
        except RuntimeError as err:
            # loading saved state dict
            if jit:
                warnings.warn(f"File {model_path} is not a JIT archive. Loading as a state dict instead")
                jit = False
            state_dict = torch.load(opened_file, map_location="cpu")

[samuela]

Same issue here with Ubuntu and CUDA 11.1 even after #227.

Update: it was a GPU driver issue. torch.cuda.is_available() was false, so device="cuda" failed.