sarif747/tinyLlama-heartfailure-education-chat

Model Card for Model ID

Model Details

Model Description

This model is a fine-tuned version of TinyLlama/TinyLlama-1.1B-Chat-v1.0, specialized for heart failure patient education and self-management guidance. It is designed to provide clear, compassionate, and evidence-based answers about symptoms, medications, lifestyle, and rehabilitation for patients and caregivers.

Developed by: Mohammad Arif Shaik

Funded by: Independent research (academic, non-commercial)

Shared by: Mohammad Arif Shaik (Hugging Face: sarif747 )

Model type: Causal Language Model (Decoder-only, chat-tuned)

Language: English

License: CC-BY-NC 4.0 (Non-commercial use)

Finetuned from: TinyLlama/TinyLlama-1.1B-Chat-v1.0

Model Sources [optional]

Repository: sarif747/tinyLlama-heartfailure-education-chat

Dataset: sarif747/heart-failure-education-qa

Base model: TinyLlama/TinyLlama-1.1B-Chat-v1.0

Direct Use

This model can be used for:

Conversational AI assistants focused on heart failure education

Generating patient-friendly explanations of heart health concepts

Supporting caregiver and patient self-management tools

Integrating into healthcare chatbots for educational content

Downstream Use [optional]

Further fine-tuning on broader cardiovascular education topics

Integration into multimodal healthcare support systems [More Information Needed]

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

Clinical diagnosis or emergency decision-making

Medical advice substitution without professional supervision

Use in high-risk healthcare settings without human oversight [More Information Needed]

Bias, Risks, and Limitations

The dataset is derived from AHA educational materials, which reflect standard U.S. health guidelines; cultural or regional variations in care may not be represented.

Model responses are educational, not prescriptive.

Potential simplifications may occur in medical terminology to improve readability.

The model may produce incomplete or inaccurate advice if prompted outside its intended context. [More Information Needed]

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations. Users and developers should:

Use the model for patient education support, not for medical decision-making.

Review generated outputs for accuracy and readability.

Disclose that the model is AI-assisted and non-clinical when deployed publicly.

How to Get Started with the Model

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

base_model = "TinyLlama/TinyLlama-1.1B-Chat-v1.0" adapter = "sarif747/tinyLlama-heartfailure-education-chat"

tokenizer = AutoTokenizer.from_pretrained(base_model, use_fast=True, trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype=torch.float16, device_map="auto") model = PeftModel.from_pretrained(model, adapter, torch_dtype=torch.float16) model.eval()

chat = pipeline("text-generation", model=model, tokenizer=tokenizer, device_map="auto")

prompt = "When is it safe for a heart failure patient to start exercising?" response = chat(prompt, max_new_tokens=150) print(response[0]["generated_text"])

Training Details

Training Data

Extracted educational content from American Heart Association (AHA) PDFs

Converted into question-answer pairs using OpenAI GPT-based prompt generation

Cleaned and validated to ensure medical accuracy and literacy alignment

Training Procedure

Base model: TinyLlama-1.1B-Chat-v1.0

Fine-tuning framework: PEFT (Parameter-Efficient Fine-Tuning) using LoRA adapters

Tokenization: TinyLlama tokenizer (SentencePiece)

Training regime:

Learning rate: 2e-4

Batch size: 64

Epochs: 3

Optimizer: AdamW

Mixed-precision (fp16)

Context length: 2048 tokens

Speeds, Sizes, Times

Model size: ~1.1B parameters

Adapter size: ~150 MB

Training duration: ~4 hours on GPU T4

Evaluation

Testing Data

A 10% held-out subset of the dataset covering:

Diet and sodium restriction Exercise recommendations Symptom monitoring and emergency response Medication adherence and fluid management

Factors

Clarity Accuracy Empathy in responses Health literacy alignment (6th–8th grade reading level)

Metrics

BLEU Score: 0.82 ROUGE-L: 0.87 Human evaluation (accuracy): 92% Human evaluation (readability): 95%

Results

The fine-tuned model demonstrates strong accuracy and patient-friendly communication. Responses are consistent, context-aware, and align with AHA educational guidance.

Metric Score BLEU 0.82 ROUGE-L 0.87 Accuracy (Human Eval) 92% Readability 95%

Summary

TinyLlama-HeartFailure-Education-Chat is a lightweight, fine-tuned conversational model for heart failure education and caregiver support. It leverages evidence-based materials and compassionate tone to improve patient understanding while maintaining low computational cost.