Failed to load BLIP about clip-interrogator HOT 10 CLOSED

It required BLIP being installed as an "egg". Thanks to @justindujardin the BLIP install no longer needs to be egg. I've updated the instructions now. Let me know if this fixes things for ya!

# install torch with GPU support for example:
pip3 install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu117

# install clip-interrogator and blip
pip install clip-interrogator==0.3.3
pip install git+https://github.com/pharmapsychotic/BLIP.git

from clip-interrogator.

Okay! Thanks so much @justindujardin and @ZetiMente install should be simplified now to just pip install clip-interrogator==0.3.5 \o/
@dante8300 can you give it a try? Everything work now for ya?

from clip-interrogator.

Thanks for the excellent project and for absorbing the package changes. I can confirm that your install instructions work without my fork now.

BTW, If you wanted to simplify to just pip installing clip-interrogator, you could probably do it like this: https://github.com/justindujardin/clip-interrogator/commit/a766f19db6ee06f31f7a80d823224dfb3dad4682

from clip-interrogator.

@justindujardin ooh finally a syntax that works with setup.py BUT ultimately didn't work with pip install 😢 ERROR: Packages installed from PyPI cannot depend on packages which are not also hosted on PyPI.
So I think options are
a) post a BLIP package to pypi like ci-blip or something
b) just move the BLIP code to sub directory of this repo and bundle up together for easy pip install.

from clip-interrogator.

@ZetiMente ooh nice find! I'll test that version out.

from clip-interrogator.

Hi @pharmapsychotic, thanks for getting back. The error has changed slightly. Currently, I have
torchvision --extra-index-url https://download.pytorch.org/whl/cu116 which matches the version I have but not sure if the diff in version is related. Is there anything else i could try?

Loading BLIP model...
Traceback (most recent call last):
  File "C:\lien\style-transfer\diff\clip-interrogator\run_cli.py", line 93, in <module>
    main()
  File "C:\ken\style-transfer\diff\clip-interrogator\run_cli.py", line 54, in main
    ci = Interrogator(config)
  File "C:\Users\ken\Anaconda2\envs\StableDiffusion\lib\site-packages\clip_interrogator\clip_interrogator.py", line 58, in __init__
    blip_model = blip_decoder(
  File "C:\Users\ken\Anaconda2\envs\StableDiffusion\lib\site-packages\blip\models\blip.py", line 175, in blip_decoder
    model,msg = load_checkpoint(model,pretrained)
  File "C:\Users\ken\Anaconda2\envs\StableDiffusion\lib\site-packages\blip\models\blip.py", line 218, in load_checkpoint
    checkpoint = torch.load(cached_file, map_location='cpu')
  File "C:\Users\ken\Anaconda2\envs\StableDiffusion\lib\site-packages\torch\serialization.py", line 777, in load
    with _open_zipfile_reader(opened_file) as opened_zipfile:
  File "C:\Users\ken\Anaconda2\envs\StableDiffusion\lib\site-packages\torch\serialization.py", line 282, in __init__
    super(_open_zipfile_reader, self).__init__(torch._C.PyTorchFileReader(name_or_buffer))
RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

from clip-interrogator.

Seems to be stuck at loading the checkpoint. Not sure what to make to this error, any advice is welcome.

This is how i am calling the client i guess nothing wrong there? python run_cli.py -i "image.PNG"

Is there a parameter to force it to use GPU ?

from clip-interrogator.

I found the parameters in the code, i think now its working. One thing i noticed was that sometimes the downloading of the BLiP_MODEL halts almost at the end, which could be why it gets corrupted. But that was just a theory.

I added the following parameters to the Interrogator Config, and it forced me to download the model again, updating the file in the process. It works, but I see that 6GB in VRAM is not enough.
from PIL import Image from clip_interrogator import Config, Interrogator image_path="sa1.PNG" image = Image.open(image_path).convert('RGB') ci = Interrogator(Config(clip_model_name="ViT-L-14/openai",blip_model_type="base",device="cuda",blip_max_length=24)) print(ci.interrogate(image))

from clip-interrogator.

@pharmapsychotic, just to let you know its solved, thanks for the help.

from clip-interrogator.

outstanding! hmm yeah looks like maybe the large model you have partially downloaded and is incomplete corrupted file.

from clip-interrogator.