ImportError: numpy.core.multiarray failed to import

[lvilaca16]

I am currently using salesforce-lavis to use BLIP-2. I am getting errors regarding incompatibilities between numpy and opencv. Despite multiple warnings, I updated both packages and it worked, thus I am listing this issue for registration purposes.

Current environment:

name: blip2
channels:
  - defaults
dependencies:
  - python=3.11
  - pip
  - pip:
    - salesforce-lavis

It installs the following versions:

• opencv-python: 4.5.5.64
• numpy: 2.0.2

Error traceback:

RuntimeError: module compiled against ABI version 0x1000009 but this version of numpy is 0x2000000                                     
Traceback (most recent call last):                                                                                                     
 File "/nas-ctm01/homes/luismsalgado/deep_av_dial/scripts/run_blip2.py", line 4, in <module>                                          
   from lavis.models import load_model_and_preprocess                                                                                 
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/lavis/__init__.py", line 15, in <module>          
   from lavis.datasets.builders import *                                                                                              
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/lavis/datasets/builders/__init__.py", line 8, in <
module>                                                                                                                                
   from lavis.datasets.builders.base_dataset_builder import load_dataset_config                                                       
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/lavis/datasets/builders/base_dataset_builder.py", 
line 18, in <module>                                                                                                                   
   from lavis.processors.base_processor import BaseProcessor                                                                     
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/lavis/processors/__init__.py", line 10, in <module
>
   from lavis.processors.alpro_processors import (
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/lavis/processors/alpro_processors.py", line 13, in
<module>
   from lavis.processors.randaugment import VideoRandomAugment
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/lavis/processors/randaugment.py", line 8, in <modu
le>
   import cv2
 File "/nas-ctm01/homes/luismsalgado/.conda/envs/blip2/lib/python3.10/site-packages/cv2/__init__.py", line 8, in <module>
   from .cv2 import *
ImportError: numpy.core.multiarray failed to import

My solution:

pip install -I -U numpy
pip install -I -U opencv

[changliu98]

Sams issue with Python 3.9.20 and numpy 2.0.2. I solved it by use Conda to create a Python 3.11 env with pip installed numpy 2.1.2, pip installed salesforce-lavis==1.0.2 and conda installed opencv, which should be 4.10.0 without specifying version number
UPDATE
when run models still got runtime error due to numpy or opencv

[xyling2023]

Sams issue with Python 3.9.20 and numpy 2.0.2. I solved it by use Conda to create a Python 3.11 env with pip installed numpy 2.1.2, pip installed salesforce-lavis==1.0.2 and conda installed opencv, which should be 4.10.0 without specifying version number

I encountered the same issue. Following your method for installation:

conda create -n lavis python=3.11
conda activate lavis
pip install numpy==2.1.2
pip install salesforce-lavis==1.0.2
conda install opencv

Then, using from lavis.models import model_zoo raises an error indicating that some packages are missing. I installed the following packages: pip install torchaudio moviepy peft easydict diffusers

However, when I continue using from lavis.models import model_zoo, I still get the error:

RuntimeError: Failed to import diffusers.models.autoencoders.autoencoder_kl because of the following error (look up to see its traceback):
Failed to import diffusers.schedulers.scheduling_dpmsolver_multistep because of the following error (look up to see its traceback):
maximum recursion depth exceeded

Have you encountered such a problem? Thank you for your help!

[changliu98]

Only installs, but sill have runtime error

When you run ``` pip install salesforce-lavis==1.0.2 ``` pip will uninstall numpy then install `numpy 2.0.2` for you, just run `pip install numpy==2.1.2` and install 2.1.2 ignoring the dependency error, I tested on my machine and it works (though i think the they need to update the dependencies). My steps is listed here: ``` conda create -n lavis python=3.11 conda activate lavis pip install salesforce-lavis==1.0.2 pip install numpy==2.1.2 conda install opencv ``` and ``` Python 3.11.10 (main, Oct 3 2024, 07:29:13) [GCC 11.2.0] on linux Type "help", "copyright", "credits" or "license" for more information. >>> from lavis.models import model_zoo /home/hello/anaconda3/envs/lavis/lib/python3.11/site-packages/fairscale/experimental/nn/offload.py:19: FutureWarning: `torch.cuda.amp.custom_fwd(args...)` is deprecated. Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead. return torch.cuda.amp.custom_fwd(orig_func) # type: ignore /home/hello/anaconda3/envs/lavis/lib/python3.11/site-packages/fairscale/experimental/nn/offload.py:30: FutureWarning: `torch.cuda.amp.custom_bwd(args...)` is deprecated. Please use `torch.amp.custom_bwd(args..., device_type='cuda')` instead. return torch.cuda.amp.custom_bwd(orig_func) # type: ignore >>> ```

When generates output still outputs runtime error due to incompatible numpy ver

[xyling2023]

Only installs, but sill have runtime error
When generates output still outputs runtime error due to incompatible numpy ver

To resolve version compatibility issues between diffusers and huggingface_hub, as well as between numpy and lavis, I applied the following fixes, and now the demo runs successfully:

pip install diffusers==0.16.0
pip install huggingface_hub==0.24.6
pip install spacy==3.6.0
pip install numpy==1.24.0

[cqray1990]

Only installs, but sill have runtime error
When generates output still outputs runtime error due to incompatible numpy ver

To resolve version compatibility issues between diffusers and huggingface_hub, as well as between numpy and lavis, I applied the following fixes, and now the demo runs successfully:

pip install diffusers==0.16.0
pip install huggingface_hub==0.24.6
pip install spacy==3.6.0
pip install numpy==1.24.0

where mode zoo models can be downloaded?

[fgy624]

You can try to install opencv-python! I solve this problem by doing this
pip install opencv-python

[KejiaZhang-Robust]

You can try the following steps to set up your environment:

conda create -n lavis python=3.11 -y  
conda activate lavis  
pip install --upgrade pip  
pip install salesforce-lavis  
pip install opencv-python 

[JCSTARS]

You can try the following steps to set up your environment:

conda create -n lavis python=3.11 -y  
conda activate lavis  
pip install --upgrade pip  
pip install salesforce-lavis  
pip install opencv-python 

Thanks a lot! I've faced the same problem, and it really works!

[Mei0211]

This question has also been troubling me for a long time. The final solution is to follow the latest version of OpenCV Python, as it supports numpy>2.0
pip install opencv-python=4.10.0.84

[KejiaZhang-Robust]

This question has also been troubling me for a long time. The final solution is to follow the latest version of OpenCV Python, as it supports numpy>2.0 pip install opencv-python=4.10.0.84

Maybe it’s a good practice to clear the pip cache to avoid potential conflicts by running before setup a new environmental.

pip cache purge