Chitrarth: Bridging Vision and Language for a Billion People
1. Introduction
Chitrarth (Chitra: Image; Artha: Meaning) is a multilingual VLM that integrates a state-of-the-art multilingual Large Language Model (LLM) with a vision module. This model is trained primarily on multilingual image-text data and is designed to work across 10 prominent Indian languages, including Hindi, Bengali, Telugu, Tamil, Marathi, Gujarati, Kannada, Malayalam, Odia, and Assamese, as well as English
2. Model Summary
Key Features
- Model: Krutrim-1 as the base LLM, SigLIP as the visual encoder with 2 layer MLP
- Languages Supported: 10 Indic languages - Hindi, Bengali, Telugu, Tamil, Marathi, Gujarati, Kannada, Malayalam, Odia, and Assamese, as well as English
- Usage: General purpose VLM
3. API Platform
Visit Chitrarth Online to access the model via the web interface.
4. Inference code
git clone https://github.com/ola-krutrim/Chitrarth.git
conda create --name chitrarth python=3.10
conda activate chitrarth
cd Chitrarth
pip install -e .
python chitrarth/inference.py --model-path "krutrim-ai-labs/chitrarth" --image-file "assets/govt_school.jpeg" --query "Explain the image. "
5. Evaluation Results
Performance against SOTA VLMs on different academic multimodal tasks. Our model consistently outperforms IDEFICS 2 (7B) and PALO 7B on different benchmarks while remaining competitive on TextVQA and Vizwiz.
We introduce BharatBench, a comprehensive evaluation benchmark suite designed for 10 under-resourced Indic languages across 3 tasks. The performance of Chitrarth on the BharatBench Evaluation framework sets a strong baseline for future research in this domain. Our model is unique in its ability to handle all included languages.
Below are the performance results of Chitrarth on BharatBench across three evaluation tasks: POPE, LLaVA-Bench, and MMVet.
| Language | POPE | LLaVA-Bench | MMVet |
|---|---|---|---|
| Telugu | 79.9 | 54.8 | 43.76 |
| Hindi | 78.68 | 51.5 | 38.85 |
| Bengali | 83.24 | 53.7 | 33.24 |
| Malayalam | 85.29 | 55.5 | 25.36 |
| Kannada | 85.52 | 58.1 | 46.19 |
| Assamese | 55.59 | 59.1 | 37.29 |
| Tamil | 83.28 | 58.3 | 34.31 |
| Marathi | 79.17 | 52.8 | 40.96 |
| Gujarati | 84.75 | 55.9 | 39.03 |
| Odia | 82.03 | 62.8 | 19.67 |
| English | 87.63 | 67.9 | 30.49 |
6. License
This code repository and the model weights are licensed under the Krutrim Community License.
7. Citation
@inproceedings{
khan2024chitrarth,
title={Chitrarth: Bridging Vision and Language for a Billion People},
author={Shaharukh Khan, Ayush Tarun, Abhinav Ravi, Ali Faraz, Praveen Kumar Pokala, Anagha Bhangare, Raja Kolla, Chandra Khatri, Shubham Agarwal},
booktitle={NeurIPS Multimodal Algorithmic Reasoning},
year={2024},
}
8. Contact
Contributions are welcome! If you have any improvements or suggestions, feel free to submit a pull request on GitHub.
9. Acknowledgement
Chitrarth is built with reference to the code of the following projects: Transformers, and LLaVA-1.5. Thanks for their awesome work!