--- 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: Ila-AI/IlaAI-v1 datasets: - KissanAI/Thinking-climate-100k pipeline_tag: text-generation ---
IlaAI Logo # IlaAI-v1.1 🌱 ### Earth · Crop · Intelligence **An open-source Agricultural AI for Bharat's Farmers** [![HuggingFace](https://img.shields.io/badge/HuggingFace-Ila--AI-yellow?logo=huggingface)](https://huggingface.co/Ila-AI) [![GitHub](https://img.shields.io/badge/GitHub-IlaAI-green?logo=github)](https://github.com/IlaAI) [![License](https://img.shields.io/badge/License-Apache%202.0-blue)](LICENSE) [![Base Model](https://img.shields.io/badge/Base-IlaAI--v1-purple)]() [![Val Loss](https://img.shields.io/badge/Val%20Loss-0.821-brightgreen)]()
--- ## 🌾 What's New in v1.1? IlaAI-v1.1 is an improved version of IlaAI-v1 with: - **5x more training data** β€” 9,000 rows vs 1,800 rows - **Better dataset** β€” KissanAI Thinking-climate-100k (chain-of-thought reasoning) - **More training** β€” 3,000 iterations vs 2,000 iterations - **Better Val Loss** β€” 0.821 vs 0.874 - **Multilingual system prompt** β€” responds in user's language --- ## πŸš€ Quick Start ```python from mlx_lm import load, generate from mlx_lm.sample_utils import make_sampler model, tokenizer = load("Ila-AI/IlaAI-v1.1") messages = [ {"role": "system", "content": "You are IlaAI, an expert agricultural assistant for Indian farmers. Always respond in the same language the user writes in. Keep answers concise, practical and under 200 words."}, {"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=1000, sampler=sampler, verbose=True) ``` --- ## πŸ“Š Training Details | Detail | Value | |--------|-------| | Base Model | IlaAI-v1 (Qwen3-4B) | | Framework | MLX LoRA | | Hardware | Apple M4 Mac Mini (24GB) | | Dataset | KissanAI/Thinking-climate-100k | | Training rows | 9,000 | | Validation rows | 1,000 | | Training iters | 3,000 | | LoRA rank | 8 | | Final Val Loss | **0.821** | | Peak Memory | 4.507 GB | --- ## πŸ—£οΈ Language Support | Language | Status | |----------|--------| | English | βœ… Excellent | | Hindi | βœ… Good | | Telugu | ⚠️ Basic | | Tamil | ⚠️ Basic | | Kannada | ⚠️ Basic | | Others | ⚠️ Basic | > Full multilingual support coming in **IlaAI-v2** with real multilingual agriculture data. --- ## ⚠️ Limitations - **Best performance in English** β€” primary training language - **Hindi** β€” good quality responses - **Other Indian languages** β€” basic support, improving in v2 - **Text only** β€” vision model coming separately --- ## πŸ—ΊοΈ Roadmap ### βœ… Phase 1 β€” Foundation - [x] IlaAI-v1 β€” English agriculture advisory - [x] IlaAI-v1.1 β€” Improved English + basic multilingual ### πŸ”„ Phase 2 β€” True Multilingual (v2) - [ ] Real multilingual agriculture data (BPCC + AI4Bharat) - [ ] 22+ Indian languages with proper quality - [ ] Release IlaAI-v2 ### πŸ‘οΈ Phase 3 β€” Vision - [ ] Crop disease detection from photos - [ ] IlaAI-Vision model ### πŸ“± Phase 4 β€” Mobile App - [ ] Free Android & iOS app - [ ] On-device inference - [ ] Voice input in Indian languages ### 🌍 Phase 5 β€” Global - [ ] Expand beyond India - [ ] Support farmers worldwide - [ ] API for developers --- ## πŸ“œ License Apache 2.0 β€” free to use, fine-tune, and build upon. --- ## πŸ™ Acknowledgements - [KissanAI](https://huggingface.co/KissanAI) for open-sourcing Dhenu models and datasets - [Qwen Team](https://huggingface.co/Qwen) for Qwen3 base models - [Apple MLX Team](https://github.com/ml-explore/mlx) for MLX framework - Every Indian farmer who inspired this project 🌾 ---
For the hands that feed a billion 🌾
HuggingFace Β· GitHub