VaidhLLaMA-3.2-3B-Instruct

VaidhLLaMA-3.2-3B-Instruct is a specialized Large Language Model fine-tuned for the domain of Ayurveda. It is built upon the Llama-3.2-3B-Instruct architecture and has been optimized to understand and reason with Ayurvedic concepts, physiology (Sharir Kriya), and clinical applications.

Model Details

• Model Name: VaidhLLaMA-3.2-3B-Instruct
• Base Model: meta-llama/Llama-3.2-3B-Instruct
• Developed By: Vivekdas
• Language: English, Hindi, Sanskrit (Domain-specific terminology)
• License: Llama 3.2 Community License
• Architecture: Transformer-based Auto-Regressive Language Model

Performance

VaidhLLaMA demonstrates strong performance on the BhashaBench-Ayur benchmark, outperforming its base model and other similarly sized models in domain-specific tasks.

Model	Accuracy (%)	Note
VaidhLLaMA-3.2-3B	41.91%	Fine-tuned Ayurveda Specialist
Llama-3.2-3B-Instruct	40.74%	Base Model
Llama-3.2-1B	27.58%	Tiny Model

Intended Use

This model is designed for:

• Answering questions related to Ayurvedic medical science.
• Explaining concepts from classical Ayurvedic texts (Samhitas).
• Assisting researchers and students in the field of Ayurveda.

Disclaimer: This model is for educational and research purposes only. It should not be used as a substitute for professional medical advice, diagnosis, or treatment.

Usage

You can run this model using the transformers library:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Vivekdas/VaidhLLaMA-3.2-3B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "system", "content": "You are VaidhLLaMA, an expert AI assistant for Ayurveda."},
    {"role": "user", "content": "Explain the concept of Tridosha in Ayurveda."}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=512,
    do_sample=True,
    temperature=0.6,
    top_p=0.9
)

response = tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True)
print(response)

Citation

If you use this model in your research, please cite:

@misc{vaidhllama2024,
  author = {Vivekdas},
  title = {VaidhLLaMA: A Fine-Tuned LLM for Ayurveda},
  year = {2024},
  publisher = {Hugging Face},
  journal = {Hugging Face Repository},
  howpublished = {\url{https://huggingface.co/Vivekdas/VaidhLLaMA-3.2-3B-Instruct}}
}