AQUA-1B/README.md

---
license: apache-2.0
tags:
- text-generation-inference
- transformers
language:
- en
base_model:
- google/gemma-3-1b-it
pipeline_tag: text-generation
---

<p align="center">
  <img src="./AQUA-1B.png" alt="AQUA-1B" width="600" style="border-radius: 6px;"/>
</p>

# Model Information

**AQUA-1B** is Kurma AI’s compact and efficient **1-billion parameter Small Language Model (SLM)**, It is the **first lightweight Aquaculture domain-specific model** purpose-built for real-time aquaculture operations involving IoT sensor data, autonomous systems, and robotic agents.

Designed for edge deployments and low-latency environments, **AQUA-1B** enables on-device decision-making, real-time alert generation, and agentic task execution. It powers intelligent aquaculture systems for Water quality monitoring, Automated feeding routines, Mobile robotic inspections across ponds, tanks, and recirculating aquaculture systems (RAS)


Learn more about [Kurma AI](https://kurma.ai/company).

---

# Key Features

- **Edge-Ready Intelligence** Optimized for low-power, real-time inference on embedded devices like Raspberry Pi, Jetson Nano, and Coral TPU.
- **Agentic Task Execution** Supports multi-step agent-based workflows such as sensor checks, feeding triggers, water exchange scheduling, and autonomous health checks using instruction-following prompts.
- **IoT-Aware Reasoning** Natively understands and reasons over sensor data inputs (e.g., temperature, pH, TDS, turbidity, DO), enabling rapid decisions in fluctuating aquaculture environments.
- **Robotic Automation Control** Designed to interact with robotic systems, including underwater, and mobile pond inspectors.
- **Autonomous Alerting Systems** Powers local alert mechanisms (via SMS, Telegram bots, or MQTT) that notify farmers when water parameters exceed safe thresholds or when interventions are required.
- **Field-Deployable Decision Engine** Enables fully autonomous operation in remote hatcheries and ponds, ensuring uninterrupted control even in offline conditions or low-connectivity zones.

---

# Training Data Highlights

- Extension worker–farmer dialogues and field advisory logs  
- FAO, ICAR, NOAA, and peer-reviewed aquaculture research  
- Synthetic Q&A from 5,000+ aquaculture-focused topics  
- Climate-resilient practices, hatchery SOPs, and water quality datasets  
- Carefully curated to support **species-specific culture** methods
- **Scale:** Trained on approximately **3 million real and synthetic Q&A pairs**, totaling around **1 billion tokens** of high-quality, domain-specific data.


---

# Model Specifications

- **Base Model**: Gemma 3 1B (by [Google DeepMind](https://deepmind.google/)) 
- **Training Tokens**: ~1 Billion
- **Released On** 4, July 2025
- **Data Volume**: 3M+ expert-verified and synthetic instructions   
- **Origin**: Made in America by [Kurma AI](https://kurma.ai/)
- **Training Technic** Model is trained via Fine-tuning using (LoRA-based) Supervised Fine-Tuning (SFT).
- **Training Infrastructure**: Trained using **8 NVIDIA H200 GPU Multi Cluster** 
Special Thanks to [Nebius](https://nebius.com/)

---

# Quickstart

Transformers (Google Colab/ jupyter)


- Install dependencies
```python
!pip install transformers accelerate
```

- Log in with your Hugging Face access token
```python
from huggingface_hub import login
```

- Import model from Huggingface
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "KurmaAI/AQUA-1B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",             # Automatically uses GPU if available
    torch_dtype=torch.float16,     # Use torch.float32 if no GPU
    trust_remote_code=True
)
```

- Test Prompt
```python
prompt = "What are the most common diseases in shrimp farming and how can they be prevented?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)
```
---

# 🙏 Acknowledgements
This project was made possible thanks to:
- [Nebius](https://nebius.com/) for providing a compute grant and access to NVIDIA H200 GPU servers, which powered the model training process.
- [Google DeepMind](https://deepmind.google/) for sharing their open-source language models, which made this project possible.
- Kurma AI research team: including aquaculture experts, machine learning engineers, data annotators, and advisors who collaborated to curate, verify, and refine the domain-specific dataset used for fine-tuning this model.

---

# ⚠️ Disclaimer, Bias & Limitations

- **Domain Bias**: The model may reflect inherent biases present in the aquaculture data sources and industry practices on which it was trained.
- **Temporal Data Limitation**: Climate and environmental recommendations are based on information available up to 2024. Users should cross-check any climate-related advice against the latest advisories (e.g., IMD or NOAA updates).
- **Potential Hallucinations**: Like all large language models, Aqua-1B may occasionally generate inaccurate or misleading responses ("hallucinations").
- **Always validate critical, regulatory, or high-impact decisions with a qualified aquaculture professional.**

---

# Citation

```bibtex
@article{narisetty2025aqua,
  title={AQUA: A Large Language Model for Aquaculture \& Fisheries},
  author={Narisetty, Praneeth and Kattamanchi, Uday Kumar Reddy and Nimma, Lohit Akshant and Karnati, Sri Ram Kaushik and Kore, Shiva Nagendra Babu and Golamari, Mounika and Nageshreddy, Tejashree},
  journal={arXiv preprint arXiv:2507.20520},
  year={2025},
  doi={10.48550/arXiv.2507.20520}
}
```