On-Device Farmer Advisory with Qwen-2.5

A reproducible, deployable pipeline for running a farmer-advisory LLM on edge devices.

This repository documents the complete engineering workflow: dataset cleaning, LoRA fine-tuning, model merging, TFLite conversion, LiteRT-LM packaging, benchmarking, key lessons learned, and deployment artifacts.

This work builds upon and extends:

Udayshankar Ravikumar . A Reproducible Pipeline for On-Device Farmer Advisory: Fine-Tuning Gemma-3n with Open KCC Data. Authorea. December 24, 2025.

DOI: 10.22541/au.176659877.79810429/v1

Navigation:

• Overview
• Training
• Conversion & Packaging
• Deployment
• Reproducibility & How to Re-run

The final packaged model artifact is available on Hugging Face at https://doi.org/10.57967/hf/7392.