Open AmirulOm opened 3 weeks ago
same issue!
I manage to change the cog file and make it work, albeit I'm not sure is the right fix for it
i need to change the cuda version to 12.1
, python version to 3.9
, torch version to 2.3.1
and torchvision to 0.18.1
Here the full cog file that worked.
# Configuration for Cog ⚙️
# Reference: https://github.com/replicate/cog/blob/main/docs/yaml.md
build:
gpu: true
cuda: "12.1"
system_packages:
- "libgl1-mesa-glx"
- "libglib2.0-0"
python_version: "3.9"
python_packages:
- "timm==0.9.2"
- "transformers==4.30.2"
- "fairscale==0.4.13"
- "pycocoevalcap==1.2"
- "torch>=2.3.1"
- "torchvision>=0.18.1"
- "Pillow==9.5.0"
- "scipy==1.10.1"
- "opencv-python==4.7.0.72"
- "addict==2.4.0"
- "yapf==0.40.0"
- "supervision==0.10.0"
- git+https://github.com/openai/CLIP.git
- ipython
predict: "predict.py:Predictor"
I'm building the cog for this project https://github.com/chenxwh/Grounded-Segment-Anything on my own server with A100 and I'm getting this error
it seems that it couldn't find image configuration for CUDA: 11.7 / Python: 3.10 / Torch: 1.13 and image now doesn't have pip install.
for context, this is the
cog.yaml
If I remove the CUDA version and the torch version. It able to build the image but there would be a version mismatch error when I run the image.
Appreciate any helps from the community