Simply clone this repository and install it as a package using pip:
$ git clone git@github.com:nicoladicicco/llm-orchestrator.git
$ cd llm-orchestrator
$ pip install -e .
The repository is structured as a data folder and self-contained scripts for running the different components of the pipeline independently. Said code will be encapsulated into the LLMInterface
class in the future.
data/
contains the LLM files, the test set, and the model outputs.llm_orchestrator/
contains Python scripts for querying the LLM interface and the output validator.planning.py
runs the planning phase of the pipeline, and saves the generated tasks in the test set folder.execution.py
runs the execution phase of the pipeline, and saves the generated data structures in the test set folder.baseline.py
runs the baseline algorithm (just LLM inference without the planning and execution phases), and saves the generated data structures in the test set folder.
To run the code, clone a Mixtral-Instruct LLM in .gguf format from here and place it in data/models/
. Feel free to experiment with other models.
You may download the paper associated with this dataset here: [insert link]