finetuned-llama-3.2-1b

🌾 Agricultural Advisory LLM — Llama-3.2-1B LoRA

A LoRA fine-tuned version of Llama-3.2-1B-Instruct for agricultural question answering and crop advisory generation. Fine-tuned using Unsloth for fast, memory-efficient training on a single GPU.

---

Model Details

---

Intended Use

✅ Direct Use

• Answering crop-specific agricultural questions (pest control, irrigation, fertilization, disease management)
• Generating advisories based on remote sensing inputs (NDVI, EVI, crop stage)
• Assisting farmers, agronomists, and agri-tech applications with decision support

✅ Downstream Use

• Embedding in agri-chatbots or mobile advisory apps
• Backend for precision agriculture dashboards
• RAG pipelines with real-time sensor data

❌ Out-of-Scope Use

• Medical, legal, or financial advice
• Non-agricultural domains
• High-stakes autonomous decision making without human review

---

Training Details

Training Data

Two datasets were mixed during fine-tuning:

1. General Agricultural Advisories — structured Q&A with fields: crop, topic, question, answer
2. Farm Sensor Data Advisories — precision agriculture records with fields: crop_type, stage, NDVI, EVI, and generated Advisory

Training Hyperparameters

• Training regime: fp16 mixed precision
• LoRA rank (r): 16
• LoRA alpha: 16
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Optimizer: AdamW (8-bit)
• Hardware: Kaggle (NVIDIA T4 / P100, single GPU)
• Framework: Unsloth + TRL SFTTrainer + PEFT 0.18.1

---

How to Get Started

With Unsloth (recommended — fastest)