josedamico/tinyllama-sugarcane-full

🌱 TinyLLaMA-Sugarcane

Welcome to the first open-source LLM fine-tuned for sugarcane production! 🧠🌾

This model is a fine-tuned version of TinyLLaMA, trained specifically on sugarcane-focused data. Developed by SciCrop as part of its commitment to open innovation in agriculture, this is one of the first domain-specific small language models (SLMs) created for the agribusiness sector.

---

🚜 Why Sugarcane?

Sugarcane is one of the most important crops in Brazil and globally — but most LLMs know very little about its specific production cycle, challenges, and terminology.

By fine-tuning TinyLLaMA on 2,000+ question/answer pairs from real-world sugarcane use cases, we aim to deliver:

• ✅ Better accuracy
• ✅ Clearer answers
• ✅ Local deployment capabilities for agricultural experts, cooperatives, and researchers

---

🔍 Model Details

• Base model: TinyLLaMA-1.1B-Chat
• Fine-tuned on: Domain-specific QA pairs related to sugarcane
• Architecture: Causal LM with LoRA + QLoRA
• Tokenizer: LLaMATokenizer
• Model size: ~1.1B parameters
• Format: Available in both HF standard and GGUF for local/Ollama use

---

🧪 Try it locally with Ollama

We believe local models are the future for privacy-sensitive, domain-specific AI.

You can run this model locally using Ollama:

ollama run infinitestack/tinyllama-sugarcane

👉 Or explore the model directly:
https://ollama.com/infinitestack/tinyllama-sugarcane

---

🌐 About InfiniteStack

This model is part of InfiniteStack, a platform by SciCrop that helps companies in the agri-food-energy-environment chain create, train, and deploy their own AI and analytics solutions — securely and at scale.

📦 InfiniteStack offers:

• A containerized platform that runs on-prem or in private cloud
• Full support for SLMs and LLMs using your real and private data
• No/Low-code interfaces to Collect, Automate, Leverage, Catalog, Observe, and Track data pipelines and AI assets

🌐 Learn more: https://infinitestack.ai

---

🧠 Why Small Language Models (SLMs)?

SLMs are great when:

• You need local inference (offline, on-device, or private)
• Your domain is narrow and specific
• You want full control over fine-tuning and usage
• You care about speed, size, and cost-efficiency

Big isn’t always better. Sometimes, smart and focused beats giant and generic. 💡

---

🤝 Community & Open Innovation

This work reflects SciCrop’s ongoing commitment to the open-source ecosystem, and to creating useful, usable AI for real-world agribusiness.

Feel free to fork, contribute, fine-tune further, or use it in your own ag project.
We’d love to hear how you're using it!

---

📂 Files included

This repo includes:

• config.json
• tokenizer.model
• tokenizer.json
• model.safetensors
• special_tokens_map.json
• generation_config.json
• tokenizer_config.json
• README.md

A merged and converted .gguf version is also available at Ollama Hub.

---

📬 Questions or Contributions?

Ping us at:
📧 info@scicrop.com
🌐 https://scicrop.com
🌱 https://infinitestack.ai

Made with ☕, 🌾 and ❤️ in Brazil
by @josedamico and the InfiniteStack team