Model type: LoRA adapter for Llama-3.2-1B-Instruct
Languages: English, Hindi
Intended use: Instruction following and conversational tasks, with improved capability on Indic (especially Hindi) data.

Training Data

Fine-tuned on a subset of:

ai4bharat/indic-instruct-data-v0.1

Mainly using the anudesh and oasst1 configs (English + Hindi splits).

How to Use

Load with Unsloth (Recommended)

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "Utkarsh736/llama_lora_indic_instruct",
    max_seq_length = 2048,
    dtype = None,
    load_in_4bit = True,
)
FastLanguageModel.for_inference(model)

Load with PEFT + Transformers

Python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(
    "unsloth/Llama-3.2-1B-Instruct",
    device_map="auto",
    load_in_4bit=True,
)

model = PeftModel.from_pretrained(base_model, "Utkarsh736/llama_lora_indic_instruct")
tokenizer = AutoTokenizer.from_pretrained("Utkarsh736/llama_lora_indic_instruct")

Acknowledgements

Training powered by Unsloth
Dataset by AI4Bharat