chigkim/VOLlama

Document build process

Closed this issue · 4 comments

ogomez92 commented

Hello,
I'm struggling a bit to build this on mac. I created a Conda environment with python 3.11 since it seems to be required by this tool, based on it looking for a file in python 3.11 when building. basically I want to build it for ARM64.

However it's looking inside a .venv folder. Do I need to create a virtual environment instead? It seems strange that there is no requirements file also.

Steps:

  1. Created conda environment.
  2. Installed requirements.txt
  3. ./build.sh

result:
'/Volumes/ssd/code/python/VOLlama']
2795 INFO: Appending 'datas' from .spec
Unable to find '/Volumes/ssd/code/python/VOLlama/.venv/Lib/python3.11/site-packages/llama_index/core/_static/nltk_cache' when adding binary and data files.
(vollama) uri@mini VOLlama %

Thanks.

chigkim commented

Yes, definitely create virtual environment with python -m venv .venv.
Also there should be requirements.txt if you cloned it.
https://github.com/chigkim/VOLlama/blob/main/requirements.txt
However, requirements.txt does not specify versions, so it won't be compatible if you build with latest dependencies unfortunately.
I have to either go back and track down which versions are compatible or make VOLllama to work with latest dependencies.

ogomez92 commented
chigkim commented

It may build fine, but some of the features won't work anymore. For example, sending image to a vision model via Ollama.

chigkim commented

I provided patch to work with latest dependencies.

pip install -U -r requirements.txt
git apply lib-win.patch

If you're on mac, use lib-mac.patch instead.