agnivamaiti/NagaLLaMA-3.2-3B-Instruct

NagaLLaMA-3.2-3B-Instruct

NagaLLaMA-3.2-3B-Instruct is a Low-Rank Adapter (LoRA) fine-tune of the Llama-3.2-3B-Instruct model, designed to understand and generate text in Nagamese (Naga Pidgin/Creole).

This model serves as a general-purpose instruction-following assistant for the Nagamese language, capable of answering queries, translating, and maintaining conversation in the local dialect used in Nagaland, India.

Model Details

• —Developer: Agniva Maiti
• —Base Model: meta-llama/Llama-3.2-3B-Instruct
• —Language: Nagamese (nag)
• —Fine-tuning Method: LoRA (Low-Rank Adaptation)
• —Precision: fp16

Training Data

The model was trained on the [NagaNLP Conversational Corpus](https://huggingface.co/datasets/agnivamaiti/naganlp-conversational-corpus), which contains 10,021 Nagamese instruction-following pairs.

Data Splitting: To ensure robust evaluation, the dataset was split as follows:

• —Training: 80% (approx. 8,000 samples)
• —Validation: 10% (used for metric evaluation during training)
• —Test: 10% (held-out for final testing)

This release represents the final model from a data-scaling ablation study, trained on 100% of the available training split.

Training Hyperparameters

• —Epochs: 3
• —Batch Size: 2 (per device) with 8 gradient accumulation steps
• —Sequence Length: 512
• —Learning Rate: 2e-4
• —LoRA Config:
• —Rank (r): 16
• —Alpha: 32
• —Dropout: 0.05
• —Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

Intended Use

This model is intended for:

• —Chatbots and assistants requiring Nagamese language support.
• —Research into low-resource language modeling for creole languages.
• —Translation assistance between English and Nagamese.

How to Use

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

# Load Base Model
base_model_id = "meta-llama/Llama-3.2-3B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Load NagaLLaMA Adapter
adapter_id = "agnivamaiti/NagaLLaMA-3.2-3B-Instruct"
model = PeftModel.from_pretrained(model, adapter_id)
tokenizer = AutoTokenizer.from_pretrained(base_model_id)

# Inference
prompt = "Machine Learning ki ase aru kote use hoi?"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(
        **inputs,
        max_new_tokens=150,      
        do_sample=True,
        temperature=0.3,
        top_k=15,              
        top_p=0.3,
        repetition_penalty=1.2,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.pad_token_id
    )

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Limitations & Safety

• —Hallucinations: Like all LLMs, this model may generate incorrect information.
• —Bias: The model inherits biases from the base Llama 3.2 model and the specific dialectal patterns found in the training data.
• —Critical Use: Not suitable for medical, legal, or financial advice.

Credits

• —Acknowledgments: Special thanks to the friends who validated the dataset and model outputs, and to RespAI Lab, KIIT for supporting the research and publication of this work.