AyurEze Llama-3B v1.0 is a LoRA-finetuned variant of unsloth/Llama-3.2-3B-Instruct specialized in Ayurvedic question answering, Sanskrit sloka interpretation, Panchakarma treatment protocols, and bilingual explanations in Tamil + English.

✨ Highlights

Retrieves and explains Charaka Samhita sutras
Provides dosage, contraindications, formulations for classical drugs
Supports Tamil herb names (உதா: நன்னாரி, கிராம்பு, ஏலக்காய்)
Can answer mixed language questions (Tamil + English + Sanskrit)
Ideal for clinics, college students, teleconsultation tools

🧠 Base Model

Model: unsloth/Llama-3.2-3B-Instruct
Quantization: 4-bit loading (bitsandbytes)
Architecture: LLaMA-3 style Transformer

🏋️ Training Details

Finetuning method: LoRA (PEFT)
Trainable parameters: ~24M (0.75%)
Epochs: 2
Training examples: ~100
Final training loss: ~2.10
Optimizer: paged_adamw_32bit
Framework: Unsloth + HuggingFace Trainer

📚 Dataset

Primary dataset: immverse-ai/Ayurveda_Dataset

Includes:

Sanskrit verses from Charaka Chikitsa Sthana
English transliteration
Clinical interpretation & treatment recommendations
Ayurvedic herb data
Panchakarma therapy explanations

🧪 Example Usage

python

from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="ayureasehealthcare/ayureze-llama-3b-1.0",
    tokenizer="ayureasehealthcare/ayureze-llama-3b-1.0"
)

print(
    pipe("Provide the Charaka reference for Sandhigatavata and explain it.")[0]["generated_text"]
)