macro-counter

Using LLMs to count macro nutrients (macros)

Setup

Tested using Nvidia GPU for acceleration. If using Nvidia, ensure you have compatible CUDA runtime (run nvidid-smi) and CUDA toolkit (run nvcc --version) installed for building llama-cpp-python.

Copy .env_example to .env and put in your API key from FoodData Central

python -m venv env
source env/bin/activate
pip install -r requirements.txt
CMAKE_ARGS="-DLLAMA_CUDA=on" pip install llama-cpp-python --force-reinstall --no-cache-dir
python download_models.py
./start_server.sh &
python main.py

TODO

• Parse recipes from the web
• Parse strings to pint quantities
• Add ability to cache recipes
• Calculate macro needs based on height, weight, age, gender
• Use JSON mode in instructor so llama-3 can be used
• Add RecipeBook object that keeps track of cached recipes
• Get macro values for each ingredient
• Adjust macros based on goals: lose weight, build muscle, maintain
• Create meal plan for individual based on recipes and macro requirements
• Create meal plan for groups of people
• Create shopping list based meal plan
• Calculate prices using grocery stores' APIs

Built using

• llama-cpp-python for serving the LLM backends, function calling
• pydantic for object orientation and typing
• instructor for generating instances of pydantic models using LLM's
• pint for unit conversion
• fooddatacentral client for USDA FoodData Central API
• USDA FoodData Central API: U.S. Department of Agriculture, Agricultural Research Service. FoodData Central, 2019. fdc.nal.usda.gov.

See Also

• Macro Counting with Python by @AdamLiscia