README.md

metadata

language:
  - en
  - hi
  - mr
  - gu
  - ta
  - ml
license: llama2
tags:
  - multilingual
  - instruction-tuning
  - llama2

RomanSetu

This was trained as part of the paper RomanSetu: Efficiently unlocking multilingual capabilities of Large Language Models via Romanization. The codebase used to train and evaluate this model can be found at https://github.com/AI4Bharat/romansetu.

Usage

Clone https://github.com/AI4Bharat/romansetu and install the required dependencies. Then download or clone this model to the same machine.

Example

from transformers import AutoTokenizer, AutoModelForCausalLM
model_path = "ai4bharat/romansetu-cpt-native-300m"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(model_path)