ayureasehealthcare/ayurze-llama-3b-1.3

🧘‍♂️ AyurEze Llama 3 — Ayurvedic Clinical + Nutrition Model (LoRA Fine-tuned)

Model Name: ayureasehealthcare/ayurze-llama-3b-meals-v2
Base Model: unsloth/Llama-3.2-3B-Instruct
Parameters: 3B (24M trainable LoRA params)
Finetuned Method: QLoRA (4-bit) using Unsloth
Domain: Ayurveda clinical guidance + Ayurvedic meals / dietetics
Status: Production-ready inference

---

✨ Overview

AyurEze-Llama is a specialized Ayurvedic reasoning model designed to provide:

• Ayurvedic dietary recommendations
• Dosha-specific meal guidance
• Seasonal diet planning (Ritucharya)
• Ingredient explanation using Dravyaguna concepts
• Herbal, culinary, and clinical usage advice
• Digestive health & Agni balancing tips

This is a LoRA finetuned version of Meta’s Llama-3 architecture optimized for low-memory inference and high-quality Ayurvedic consultation.

---

📚 Training Data Sources

1. jaychedaa/Ayurveda-LLM-dataset

• Ayurvedic Q/A
• Dosha, Agni, Mala, Dhatu fundamentals
• Treatment principles (Chikitsa Siddhanta)
• Sanskrit & textual reference content

2. nadakandrew/ayurvedicmeals

• Meal descriptions and Ayurvedic dietary content
• Seasonal foods
• Ingredients and benefits
• Dosha balancing recipes

---

⚙️ Training Details

• Method: QLoRA — 4-bit training using Unsloth
• Frameworks: TRL, PEFT, HuggingFace Datasets
• Hardware: Google Colab (1x T4 GPU)
• Epochs: 2 (continue finetuning)
• Batch size: 2
• Grad accumulation: 4
• Learning rate: 5e-5
• Trainable parameters: ~24.3M
• Loss: Final training loss ≈ 0.28

Special Notes

• Gradient checkpointing was disabled for stability
• Pad token = EOS token
• Input formatted using chat templates

---

🎯 Intended Use

This model is suitable for:

• Ayurvedic diet chatbots
• Meal recommendation systems
• Wellness companion apps
• Ayurvedic nutrition consultation
• Restaurant menu intelligence
• Seasonal diet planners
• Panchakarma dietary guidance

---

🚫 Limitations & Disclaimer

⚠️ This model does not replace a licensed Ayurvedic physician.
It is for educational & informational purposes only.

• May generate incorrect medical claims
• Should not be used for diagnosis or prescription
• Always consult a BAMS doctor / registered practitioner

---

🧪 Inference (Quick Start)

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "ayureasehealthcare/ayurze-llama-3b-meals-v2"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

prompt = "Suggest an Ayurvedic dinner for Vata imbalance in winter."

inputs = tokenizer(
    prompt,
    return_tensors="pt"
).to("cuda")

output = model.generate(
    **inputs,
    max_new_tokens=300,
    temperature=0.7,
    top_p=0.9,
    pad_token_id=tokenizer.eos_token_id,
)

print(tokenizer.decode(output[0]))