Quantization made by Richard Erkhov.

[Github](https://github.com/RichardErkhov)

[Discord](https://discord.gg/pvy7H8DZMG)

[Request more models](https://github.com/RichardErkhov/quant_request)


Dhenu2-In-Llama3.2-1B-Instruct - AWQ
- Model creator: https://huggingface.co/KissanAI/
- Original model: https://huggingface.co/KissanAI/Dhenu2-In-Llama3.2-1B-Instruct/




Original model description:
---
library_name: transformers
tags:
- llama3.2 1B
---

![image/png](https://cdn-uploads.huggingface.co/production/uploads/616d9181c3bac80637586601/BejxZ9NVc9d27SsTp4LcY.png)

# Dhenu2 India 1B

## Model Overview

**Model Name:** Llama3.2-Dhenu2-In-1B-Instruct

**Architecture:** Llama3.2

**Parameters:** 1 Billion

**Release Date:** 24th October, 2024

**License:** [Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE)


## Description

Designed for efficiency, Dhenu2 India 1B provides swift agricultural insights and is optimized for deployment on resource-constrained devices. Built on the Llama3.2 architecture with 1 billion parameters, this lightweight model ensures rapid responses, making it perfect for on-device applications and mobile advisory tools used by farmers and agricultural workers.

## Intended Use

- **Mobile Applications:** Embed Dhenu2 India 1B in mobile apps to provide farmers with real-time assistance and insights directly on their smartphones.
- **On-Device Advisory Tools:** Develop lightweight advisory systems that operate efficiently on limited hardware resources.
- **Field Operations:** Utilize in-field devices for immediate agricultural support without the need for constant internet connectivity.

## Training Data

Dhenu2 India 1B was trained on a specialized dataset that reflects the varied landscape of Indian agriculture, including:
- **Instruction Set:** Over 1.5 million instructions from real and synthetic conversations.
- **Synthetic Instructions:** Generated through sophisticated pipelines to ensure comprehensive coverage of more than 4,000 agricultural topics.
- **Data Sources:** Mobile extension service logs, farmer feedback, agricultural package of practices, and localized studies.

## Training Procedure

- **Techniques:** Employed full fine-tuning to optimize the model’s performance while ensuring resource efficiency.
- **Hardware:** Trained using multi-GPU setups with NVIDIA A100 GPUs, utilizing DeepSpeed for distributed training and memory management.
- **Optimization:** Implemented flash attention mechanisms to enhance computational efficiency and reduce memory consumption, enabling seamless deployment on mobile devices.

## Evaluation

Dhenu2 India 1B was evaluated to ensure its effectiveness and efficiency for on-device applications:
- **Human Evaluation:** Tested by agricultural professionals for relevance, speed, and accuracy of responses.
- **Synthetic Evaluation:** Conducted peer assessments using other LLMs to validate consistency and correctness.
- **Performance Metrics:** Assessed based on response time, accuracy in delivering insights, and efficiency in resource usage.

## Limitations

While Dhenu2 India 1B excels in efficiency and speed, it may have limited depth compared to larger models. It is optimized for quick insights and may not handle highly complex or detailed agricultural queries as effectively as its larger counterparts.

## API

Use our platform [Dhenu](https://dhenu.ai) with a generous free quota to start building your agriculture applications.

## A note of gratitude

We want to thank our partners Microsoft and Microsoft for Startups for landing us compute. We would also like to thank our partner, Meta, for the open-source Llama models.

## Contact

For more information, support, or collaboration inquiries, please contact us at [support@kissan.ai].