🪐 spaCy Project: Prodigy Address Extraction model bootstrapped with LLM's

This project creates an address extraction model. To improve annotation efficiency, we'll experiment with using LLM's to speed up the development process.

📋 project.yml

The project.yml defines the data assets required by the project, as well as the available commands and workflows. For details, see the spaCy projects documentation.

⏯ Commands

The following commands are defined by the project. They can be executed using spacy project run [name]. Commands are only re-run if their inputs have changed.

Command	Description
install	Install packages
clean	Remove intermediate files
clean-venv	Remove the virtual environment
generate-data	Create synthetic data from LLM
ner-manual-train	NER manual annotate for training from generated (synthetic) data
ner-manual-eval	NER manual annotate for evaluation from generated (synthetic) data
ner-train-curve	NER correct annotate for training from generated (synthetic) data
ner-correct	NER correct annotate for training from generated (synthetic) data
data-merge	Merge manual and correct data for training data
ner-data-to-spacy	Convert training and evaluations to spaCy binary data
ner-data-debug	Run data debug on training and evaluation data
train	Train pipeline models
load-annotations	Load training and evaluation data as Prodigy datasets
train-vectors	Train pipeline models with vectors
evaluate	Evaluate the model and export metrics
package	Package the trained model as a pip package
visualize-model	Visualize the model's output interactively using Streamlit
document	Export README for project details

⏭ Workflows

The following workflows are defined by the project. They can be executed using spacy project run [name] and will run the specified commands in order. Commands are only re-run if their inputs have changed.

Workflow	Steps
all	install → load-annotations → ner-data-to-spacy → train-vectors → evaluate
visualize	package → visualize-model

🗂 Assets

The following assets are defined by the project. They can be fetched by running spacy project assets in the project directory.

File	Source	Description
assets/addresses.jsonl	Local	LLM-generated (synthetic) data
assets/addresses_train.jsonl	Local	Annotated training data from LLM-generated (synthetic) data
assets/addresses_eval.jsonl	Local	Annotated evaluation data from LLM-generated (synthetic) data