This PR primarily seeks to solve several setup issues that have been brought to our attention and are currently plaguing the main branch.
Issues fixed:
entrypoint.sh permissions
communication between the front / backend
seamless switch between docker and local environment
no need to set a test file if the .env if test_mode stays off
Core changes:
We are now using the llama.cpp integration and flow are our default option
The guidance server is not built anymore with docker - although this might very well change in the future
Some limitations:
GPU offloading for llama.cpp does not seem to be available in docker. This does not prevent the program to run but makes it dramatically slower. As a result, I personally recommend the local env installation if you have a gpu you want to take advantage off with BrainChulo.
More a side note than a limitation, but do not forget to change your model(s) path(s) when/if you switch from docker to local env.
Next steps:
From a core functionality perspective, the first goal is still to make the process faster and more consistent. - ideally with way smaller models than we are using now.
This PR primarily seeks to solve several setup issues that have been brought to our attention and are currently plaguing the main branch.
Issues fixed:
Core changes:
Some limitations:
Next steps: