MILU: A Multi-task Indic Language Understanding Benchmark

Overview

MILU (Multi-task Indic Language Understanding Benchmark) is a comprehensive evaluation dataset designed to assess the performance of Large Language Models (LLMs) across 11 Indic languages. It spans 8 domains and 42 subjects, reflecting both general and culturally specific knowledge from India.

This repository contains code for evaluating language models on the MILU benchmark using the lm-eval-harness framework.

Usage

Prerequisites

• Python 3.7+
• lm-eval-harness library
• HuggingFace Transformers
• vLLM (optional, for faster inference)

1. Clone this repository:

git clone --depth 1 https://github.com/AI4Bharat/MILU.git
cd MILU
pip install -e .

2. Request access to the HuggingFace 🤗 dataset here.

3. Set up your environment variables:

export HF_HOME=/path/to/HF_CACHE/if/needed
export HF_TOKEN=YOUR_HUGGINGFACE_TOKEN

The following languages are supported for MILU:

• Bengali
• English
• Gujarati
• Hindi
• Kannada
• Malayalam
• Marathi
• Odia
• Punjabi
• Tamil
• Telugu

HuggingFace Evaluation

For HuggingFace models, you may use the following sample command:

lm_eval --model hf \
    --model_args 'pretrained=google/gemma-2-27b-it,temperature=0.0,top_p=1.0,parallelize=True' \
    --tasks milu \
    --batch_size auto:40 \  
    --log_samples \
    --output_path $EVAL_OUTPUT_PATH \
    --max_batch_size 64 \
    --num_fewshot 5 \
    --apply_chat_template

vLLM Evaluation

For vLLM-compatible models, you may use the following sample command:

lm_eval --model vllm \
    --model_args 'pretrained=meta-llama/Llama-3.2-3B,tensor_parallel_size=$N_GPUS' \
    --gen_kwargs 'temperature=0.0,top_p=1.0' \
    --tasks milu \
    --batch_size auto \
    --log_samples \
    --output_path $EVAL_OUTPUT_PATH

Single Language Evaluation

To evaluate your Model on a specific language, modify the --tasks parameter:

--tasks milu_English

Replace English with the available language (e.g., Odia, Hindi, etc.).

Evaluation Tips & Observations

1. Make sure to use --apply_chat_template for Instruction-fine-tuned models, to format the prompt correctly.
2. vLLM generally works better with Llama models, while Gemma models work better with HuggingFace.
3. If vLLM encounters out-of-memory errors, try reducing max_gpu_utilization else switch to HuggingFace.
4. For HuggingFace, use --batch_size=auto:<n_batch_resize_tries> to re-select the batch size multiple times.
5. When using vLLM, pass generation kwargs in the --gen_kwargs flag. For HuggingFace, include them in model_args.

Key Features

• 11 Indian Languages: Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, and English
• Domains: 8 diverse domains including Arts & Humanities, Social Sciences, STEM, and more
• Subjects: 42 subjects covering a wide range of topics
• Questions: ~85,000 multiple-choice questions
• Cultural Relevance: Incorporates India-specific knowledge from regional and state-level examinations

Dataset Statistics

Language	Total Questions	Translated Questions	Avg Words Per Question
Bengali	7138	1601	15.72
Gujarati	5327	2755	16.69
Hindi	15450	115	20.63
Kannada	6734	1522	12.83
Malayalam	4670	1534	12.82
Marathi	7424	1235	18.8
Odia	5025	1452	15.63
Punjabi	4363	2341	19.9
Tamil	7059	1524	13.32
Telugu	7847	1298	16.13
English	14036	-	22.01
Total	85073	15377	16.77 (avg)

Dataset Structure

Test Set

The test set consists of the MILU (Multi-task Indic Language Understanding) benchmark, which contains approximately 85,000 multiple-choice questions across 11 Indic languages.

Validation Set

The dataset includes a separate validation set of 9,157 samples that can be used for few-shot examples during evaluation. This validation set was created by sampling questions from each of the 42 subjects.

Subjects spanning MILU

Domain	Subjects
Arts & Humanities	Architecture and Design, Arts and Culture, Education, History, Language Studies, Literature and Linguistics, Media and Communication, Music and Performing Arts, Religion and Spirituality
Business Studies	Business and Management, Economics, Finance and Investment
Engineering & Tech	Energy and Power, Engineering, Information Technology, Materials Science, Technology and Innovation, Transportation and Logistics
Environmental Sciences	Agriculture, Earth Sciences, Environmental Science, Geography
Health & Medicine	Food Science, Health and Medicine
Law & Governance	Defense and Security, Ethics and Human Rights, Law and Ethics, Politics and Governance
Math and Sciences	Astronomy and Astrophysics, Biology, Chemistry, Computer Science, Logical Reasoning, Mathematics, Physics
Social Sciences	Anthropology, International Relations, Psychology, Public Administration, Social Welfare and Development, Sociology, Sports and Recreation

Citation

If you use MILU in your work, please cite us:

@article{verma2024milu,
  title   = {MILU: A Multi-task Indic Language Understanding Benchmark},
  author  = {Sshubam Verma and Mohammed Safi Ur Rahman Khan and Vishwajeet Kumar and Rudra Murthy and Jaydeep Sen},
  year    = {2024},
  journal = {arXiv preprint arXiv: 2411.02538}
}

License

This dataset is released under the CC BY 4.0.

Contact

For any questions or feedback, please contact:

• Sshubam Verma (sshubamverma@ai4bharat.org)
• Mohammed Safi Ur Rahman Khan (safikhan@ai4bharat.org)
• Rudra Murthy (rmurthyv@in.ibm.com)
• Vishwajeet Kumar (vishk024@in.ibm.com)

Links

• GitHub Repository 💻
• Paper 📄
• Hugging Face Dataset 🤗