Triplet Extraction Notebook

A hands-on Jupyter notebook for extracting subject–predicate–object triplets from sentences using large language models (LLMs).

Filename: FinalTripletGenerator.ipynb

This single notebook contains all code to:

Load your sentences (from a CSV, text file, or inline list). Configure and call the LLM - Gemma3 4B - through HuggingFace, and quantize the model to 4bit using Bitsandbytes. Export extracted tripletsto CSV or JSON.

Dependencies: Transformers PyTorch TQDM Huggingface Hub

Requirements: Python 3.8+ Jupyter Notebooks An NVIDIA GPU is recommended - preferably one with support for bfloat16 (ie, Ada Lovelace and newer) for optimal performance.

Contributing and Feedback: Please suggest improvements to the prompting, or the workflow! You can contact me through ramsundar289@gmail.com

Notes: Gemma3 4B can run on GPUs like the GTX 1050 4GB, but throughput is very low. This notebook was tested on: Pop!_OS + GTX 1050 -- batch size 8 Windows 11 + RTX 3050Ti -- batch size 8 H100 server environments -- batch size 256 Performance scaled well, and for best results the batch_size can be adjusted.