@n1hility @russellb
The workaround in this PR is problematic. It doesn't use the current requirements.txt but the version from the main branch. This is likely to cause issues when we create a stable branch and modify requirements in main. Also pip install -r ... will install all dependencies in the requirements file.

I'm using this approach in my container files, which I believe is better:

1. copy requirements file from context into the container
2. convert the requirements file to a constraints file by stripping off optional dependencies
3. pip install with -c instead of -r

COPY requirements.txt /tmp
RUN sed 's/\[.*\]//' /tmp/requirements.txt >/tmp/constraints.txt
RUN CMAKE_ARGS="-DLLAMA_CUBLAS=on" python3.11 -m pip install -c /tmp/constraints.txt --force-reinstall --no-cache-dir llama-cpp-python

Thanks @tiran -- good feedback. you're right.

However, this Containerfile isn't building instructlab from the current source tree. It's explicitly pulling stable from git, so the added line also pulls requirements.txt from stable.

I do think this Containerfile should be enhanced to allow building from the current source tree, as well.