MahaBERT / IndicBERT Evaluation Benchmarks

MH Specific

Evaluation results and benchmark scores for MahaBERT (L3Cube) and IndicBERT (AI4Bharat) models on Marathi NLU tasks including sentiment, NER, and text classification

Build a Marathi model comparison framework using MahaBERT/IndicBERT benchmarks to guide model selection.

Homepage HuggingFace GitHub

Quick Start

from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained('l3cube-pune/marathi-bert-v2')
model = AutoModel.from_pretrained('l3cube-pune/marathi-bert-v2')
inputs = tokenizer('मराठी भाषा', return_tensors='pt')
print(f'Output shape: {model(**inputs).last_hidden_state.shape}')

Modality

Benchmarks (tables)

Size

Model evaluation data

License

Open Research

Format

Various

Language

Update Frequency

static

Organization

L3Cube, Pune

Schema

Field	Type	Description
model	string	Model name being evaluated
task	string	Benchmark task name
metric	string	Evaluation metric (F1, accuracy, etc.)
score	float	Benchmark score

Build With This

Create an automated Marathi model benchmarking pipeline that evaluates new models on standard tasks

Develop a cost-performance analyzer comparing Marathi model accuracy against inference latency and model size

Build a task-specific model recommender that suggests the best Marathi BERT variant for each NLP application

AI Use Cases

Model selection for Marathi taskstransfer learning baselinearchitecture comparison

Related Datasets

BiasShades Marathi (LLM Bias Evaluation)

text

FLORES-200 Benchmark

Text (parallel, Marathi)

Last verified: 2026-03-07