finetuned-Qwen2.5-1.5B-Instruct

🌾 Agricultural Advisory LLM — Qwen2.5-1.5B LoRA

A LoRA fine-tuned version of Qwen2.5-1.5B-Instruct for agricultural question answering and crop advisory generation. The smallest and fastest model in this series — ideal for edge deployment, mobile apps, or low-resource environments while still delivering solid agronomic advice.

Fine-tuned using Unsloth for fast, memory-efficient training on a single GPU.

Also available in this series:

• Llama-3.2-1B variant
• Llama-3.2-3B variant

---

Model Details

---

Training Results

---

Example Outputs

🌾 Wheat — Nitrogen Deficiency

Q: What are the signs of nitrogen deficiency?

A: Pale yellow leaves with green veins in spring wheat. Symptoms appear first at leaf tips and margins.

---

🌾 Rice — Irrigation

Q: How often should I irrigate during the tillering stage?

A: Every 7-10 days depending on temperature and soil type. Over-irrigation causes anaerobic conditions that reduce tillers by 20-30%. Under-irrigation stops tillering early. Find the optimal balance for your field.

---

🌽 Maize — NDVI Interpretation

Q: My NDVI reading is 0.42, is that healthy?

A: Normal crop condition. No action required.

---

🌿 Cotton — Pest Management

Q: What pesticide do you recommend for whitefly infestation?

A: Imidacloprid or neonicotinoid-based sprays; apply when nymphs are present to prevent resistance development. Apply in the evening if possible, avoid spraying when temperatures exceed 35°C, and consider biological control methods (parasitic wasps, predatory beetles) before resorting to chemicals. Always check recommendations from your local agricultural extension department.

---

Intended Use

✅ Direct Use

• Answering crop-specific agricultural questions (pest control, irrigation, fertilization, disease management)
• Generating advisories from remote sensing inputs (NDVI, EVI, crop growth stage)
• Decision support for farmers, agronomists, and precision agriculture systems

✅ Downstream Use

• Lightweight backend for agri-advisory chatbots and mobile apps
• Edge deployment on low-resource devices
• RAG pipelines augmented with real-time field sensor data

❌ Out-of-Scope Use

• Medical, legal, or financial advice
• Non-agricultural domains
• Autonomous high-stakes decisions without human expert 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

---

How to Get Started

With Unsloth (recommended)