Khyatimirani/pcos-fertility-llama3-8b-lora

PCOS Fertility Assistant – LoRA Fine-tuned Model

Model Overview

The PCOS Fertility Assistant is a domain-specialized language model fine-tuned to provide safe, evidence-based, and patient-friendly guidance for individuals navigating Polycystic Ovary Syndrome (PCOS) and fertility care.

This model is designed to prioritize clinical safety, grounded responses, and calm communication suitable for healthcare support environments.

⚠️ Important: This model is not a substitute for professional medical advice.

---

Model Details

• Developed by: Khyati Mirani
• Model type: LoRA fine-tuned large language model
• Base model: Meta-Llama-3.1-8B-Instruct
• Domain: Fertility, PCOS, reproductive health
• Language: English
• License: Llama 3 Community License (inherits base model license)
• Finetuning method: Unsloth LoRA

---

Intended Use

Primary Use Cases

This model is intended for:

✅ AI fertility companions
✅ IVF patient education tools
✅ Healthcare support assistants
✅ Retrieval-augmented medical QA systems
✅ Doctor-supervised patient communication workflows

The model is optimized for:

• Short, calm, medically responsible answers
• Reduced hallucination risk
• Conservative guidance
• Encouraging consultation with healthcare professionals

---

Out-of-Scope Use

This model must NOT be used for:

❌ Medical diagnosis
❌ Treatment recommendations
❌ Medication prescribing
❌ Emergency decision-making
❌ Autonomous clinical systems

Human medical oversight is strongly recommended for any real-world deployment.

---

Safety Philosophy

This model was fine-tuned with a safety-first approach:

• Avoids speculative medical claims
• Uses conservative language
• Encourages doctor consultation when appropriate
• Reduces overconfident responses

However, like all language models, it may produce incorrect or incomplete information.

Clinical validation is required before production healthcare use.

---

Training Data

The model was fine-tuned on a curated domain dataset focused on:

• PCOS fundamentals
• Fertility care concepts
• IVF-related patient questions
• Symptom awareness
• Lifestyle guidance

The dataset emphasizes structured medical knowledge and deterministic answers rather than open-ended internet text.

(Dataset is currently private.)

---

Training Procedure

• Framework: Unsloth
• Technique: Parameter-efficient fine-tuning (LoRA)
• Precision: Mixed precision (bf16/fp16 depending on hardware)
• Objective: Improve domain reliability while preserving base reasoning ability

Key Hyperparameters

Parameter	Value
Learning rate	~3e-5
Training method	LoRA
Sequence length	2048
Epochs	~3
Optimizer	AdamW 8-bit

---

How to Use

Load with Unsloth

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="Khyatimirani/pcos-fertility-assistant",
    max_seq_length=2048,
    load_in_4bit=True,
)