agnivamaiti/KokLLaMA-3.2-3B-Instruct

This is KokLLaMA v2, a fine-tuned version of Llama 3.2 3B optimized for the Kokborok language.

Model Details

• Base Model: meta-llama/Llama-3.2-3B-Instruct
• Training Method: QLoRA (Rank=32, Alpha=64)
• Target Modules: All Linear Layers (Knowledge + Syntax)
• Dataset: Cleaned Kokborok-English Instruction pairs

How to Use

from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = "meta-llama/Llama-3.2-3B-Instruct"
adapter_model = "agnivamaiti/KokLLaMA-3.2-3B-Instruct"

# Load Base
model = AutoModelForCausalLM.from_pretrained(base_model, device_map="auto")
# Load Adapter
model = PeftModel.from_pretrained(model, adapter_model)
tokenizer = AutoTokenizer.from_pretrained(adapter_model)

inputs = tokenizer("Kokborok language hwnwi tamo?", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0]))