Installation

Everything in this project — every script and every notebook — is intended to run inside the weather-doc-extractor Conda environment. Setting that up is the only prerequisite for reading and running the workflow notebooks.

Prerequisites

• Miniconda or Anaconda

• Git

• A local machine is enough for orchestration and validation. The heavy work — extraction and fine-tuning — runs on GPUs via Azure ML; see that page for the compute setup.

1 — Clone the repository

git clone https://github.com/Philip-Brohan-MO/Auto-Daily-Rainfall-MO.git
cd Auto-Daily-Rainfall-MO

2 — Create the Conda environment

The environment.yml file pins all dependencies including PyTorch, Transformers, TRL, and PEFT:

conda env create -f environment.yml
conda activate weather-doc-extractor

Important

Activate weather-doc-extractor before running any script or notebook. This is non-negotiable for reproducibility — the notebooks assume this environment.

3 — Install the package in editable mode

The Conda environment already runs pip install -e . as part of its post-link step. If you need to reinstall manually:

pip install -e .

This makes the weather-extract command available on your PATH.

4 — Verify the installation

weather-extract info

You should see a JSON summary of the project configuration.

Optional: training dependencies

The base install is inference-only. To enable fine-tuning, install the train extras:

pip install -e ".[train]"

This adds accelerate, datasets, peft, torch, transformers, and trl.

Updating

git pull
conda env update -f environment.yml --prune