The setup for GPT4RoI repo is complicated and can become a large barrier for others to contribute.
Solution
Use .devcontainer to standardize development environment and automate installation steps
Automates (see .devcontainer/postCreateCommand.sh)
Guarantees correct versions of python, cuda, cuda compiler and torch to allow building the packages using native C++ compilation such as flash-attn and mmcv
Installation of root package
Downloading of pretrained models llama-7b-hf and GPT4RoI-7B-delta-V0 and run script to apply delta
Installation of NVM, install node, install pnpm, clone gradio-box, and build package for usage in demo
Issue
The setup for GPT4RoI repo is complicated and can become a large barrier for others to contribute.
Solution
Use .devcontainer to standardize development environment and automate installation steps
Automates (see .devcontainer/postCreateCommand.sh)
flash-attn
andmmcv
llama-7b-hf
andGPT4RoI-7B-delta-V0
and run script to apply deltaVideo
I likely won't be able to work more on this, but I thought it could be useful for others.