README.md

metadata

language:
  - en
  - hi
  - te
  - ta
  - kn
  - mr
  - bn
  - gu
  - pa
license: apache-2.0
library_name: mlx
tags:
  - agriculture
  - indian-farming
  - fine-tuned
  - lora
  - qwen3
  - mlx
  - crops
  - farming
  - india
base_model: Qwen/Qwen3-4B
datasets:
  - KissanAI/india-climate-qa-synth-v1
pipeline_tag: text-generation

IlaAI-v1 🌱

Earth · Crop · Intelligence

An open-source Agricultural AI for Bharat's Farmers

---

🌾 About IlaAI

IlaAI (इला — Sanskrit for "the earth that gives") is an open-source LLM fine-tuned specifically for Indian agriculture. Built on Qwen3-4B and trained on 96,000+ Indian agriculture Q&A pairs, IlaAI helps farmers and developers get accurate, practical farming advice.

"For the hands that feed a billion 🌾"

---

✨ Capabilities

• 🌾 iagnosis** — symptoms, causes, treatments
• 💊 Pesticide & Fertilizer Advice — what to use, how much, when
• 🌦️ Weather-based Advisory — sowing, irrigation, harvest timing
• t Schemes** — eligibility, documents, how to apply
• 📈 Market Price Guidance — mandi prices, sell/hold advice
• 🐛 Pest Management — detection and treatment
• 🌱 Soil Health — soil types, nutrients, improvement tips

---

🚀 Quick Start

from mlx_lm import load, generate
from mlx_lm.sample_utils import make_sampler

model, tokenizer = load("Ila-AI/IlaAI-v1")

messages = [
    {"rolontent": "You are IlaAI, an expert agricultural assistant for Indian farmers. Answer clearly and helpfully."},
    {"role": "user", "content": "My wheat crop has yellow spots on leaves. What should I do?"}
]

text = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=False
)

sampler = make_sampler(temp=0.7, top_p=0.9)
response = generate(model, tokenizer, prompt=text, max_tokens=300, sampler=sampler, verbose=True)

---

📊 Training Details

Detail	Value
Base Model	Qwen3-4B (4-bit quantized)
Framework	MLX LoRA
Hardware	Apple M4 Mac Mini (24GB)
Dataset	KissanAI/india-climate-qa-synth-v1
Training rows	92,060
Validation rows	4,845
Training iters	5,000
LoRA rank	8
Final Val Loss	0.695
Peak Memory	3.894 GB

---

⚠️ Limitations

• Best performance in E — Hindi and other Indian languages are supported but quality varies
• Text only — does not process images (vision model coming in v2)
• v1 release — multilingual improvement planned for v2

---

🗺️ Roadmap

✅ Phase 1 — Foundation (Current)

• Organization setup (GitHub + HuggingFace)
• Dataset curation (96K rows)
• IlaAI-v1 — English agriculture advisory
• Published on HuggingFace

🔄 Phase 2 — Multilingual (Coming Soon)

• Add Hindi, Telugu, Tamil, Kannada, Marathi, Bengali, Gujarati, Punjabi datasets
• Fine-tune IlaAI-v2 on 22+ Indian languages
• Improve response quality across all languages
• Release IlaAI-v2 on HuggingFace

👁️ Phase 3 — Vision (Coming Soon)

• Fine-tune vision model on Indian crop disease images
• Support all major Indian crops — Rice, Wheat, Cotton, Sugarcane, Pulses
• Release IlaAI-Vision on HuggingFace
• Combine text + vision into one unified model

📱 Phase 4 — Mobile App (Coming Soon)

• IlaAI Android app — free, forever
• IlaAI iOS app — free, forever
• On-device inference (no internet needed)
• Voice input in Indian languages
• Available on Play Store & App Store

🌐 Phase 5 — Community (Coming Soon)

• Open dataset contributions from farmers
• Regional crop data by state
• API for developers to build on IlaAI
• WhatsApp bot integration

---

📜 License

Apache 2.0 — free to use, fine-tune, and build upon.

---

🙏 Acknowledgements

• [Kface.co/KissanAI) for open-sourcing Dhenu models and datasets
• Qwen Team for Qwen3 base models
• Apple MLX Team for MLX framework
• Every Indian farmer who inspired this project 🌾

---

For the hands that feed a billion 🌾
HuggingFace · GitHub