HiddenPractical Assignment 3: Agents
You need to develop a solution to answer questions about the composition of dishes (for example, calories and macronutrients or by recipes), using large language models (LLMs) with support for tool calling. Participants are required to use an open LLM from Hugging Face together with external APIs (a Recipe API and a Nutrition API) to accurately determine the nutritional value per serving for given dishes.
Prediction requirements
- For questions related to calories, predictions must be given in kilocalories (kcal).
- For all other nutrient parameters (protein, fat, carbohydrates, etc.) predictions must be given in grams (g).
- Round predictions up to 2 decimal places.
Constraints
- You may use any models, but model size must not exceed 4B parameters and the weights must be open.
- You may not use any additional data beyond what is already provided in the competition.
- A full run of the solution should be fit within Colab or Kaggle resource limits.
- You are allowed to use additional tools.
Notes: use the Recipe API https://api-ninjas.com/ and Nutrition API https://calorieninjas.com/api (you need to register and get the free key) together with the chosen open LLM (Hugging Face) and implement tool-calling to fetch recipe and nutrition data so you can compute accurate per‑serving values.
You can start from starting_solution.ipynb
Our website uses cookies, including web analytics services. By using the website, you consent to the processing of personal data using cookies. You can find out more about the processing of personal data in the Privacy policy