lisa-on-cuda 1.4.0

LISA (Reasoning Segmentation via Large Language Model) on cuda, now with huggingface ZeroGPU support!

---

title: lisa + gradio + fastapi + ZeroGPU emoji: ⚡ colorFrom: red colorTo: purple sdk: gradio sdk_version: 4.40.0 app_file: app.py pinned: true

LISA (Reasoning Segmentation via Large Language Model) on cuda, now with huggingface ZeroGPU support!

Update requirements.txt, needed on HuggingFace demo

python -m samgis_core.utilities.update_requirements_txt \
  --req_no_version_path requirements_no_versions.txt \
  --req_output_path requirements.txt

Pay attention about any unnecessary dependency, e.g. gradio_client==1.2.0 (it's installed with gradio).

Exec jupyter on the remote server with port forwarding on localhost

1. checkout repo, install venv with jupyter
2. port forwarding in localhost wiht private key: ssh -i /path/to/private_key [e-mail] -L 8889:localhost:8889 -N -f
3. start the jupyter-lab server
4. connect to page in localhost

Commands to work on remote virtual machines (e.g. SaturnCloud) after clone and git lfs install

cd ~/workspace/lisa-on-cuda/
rm -rf lisa_venv 
python3 -m venv lisa_venv
ln -s lisa_venv/ venv
source  venv/bin/activate
pip --version
which python
python -m pip install pip wheel --upgrade
python -m pip install pytest pytest-cov jupyterlab
python -m pip install -r requirements.txt
nohup jupyter-lab &
tail -F nohup.out

Jupyterlab Howto

To run the test.ipynb notebook you should already:

• cloned project https://huggingface.co/spaces/aletrn/lisa-on-cuda with active git lfs
• created and activated a virtualenv
• installed jupyterlab dependencies from requirements_jupyter.txt
• installed dependencies from requirements.txt

Hardware requirements for local usage

• an nvidia gpu with 10 or 12GB of memory (a T4 should suffice)
• at least 16GB of system ram

Hardware requirements on huggingface ZeroGPU

Right now (July 2024) huggingface let use ZeroGPU Nvidia A100 GPUs.

See LISA for details on the original project. Note that the authors don't keep the project updated anymore.