Chakster/Tinyllama_HeartFailure

Model Card: TinyLlama Heart Failure Optimized

Model Details

Model Name: TinyLlama Heart Failure Model Type: Fine-tuned Language Model
Base Model: TinyLlama
Domain: Heart Failure Q&A
Version: 1.0
Date: July 13, 2025

Model Description

This model is a domain-specific fine-tuned version of TinyLlama, optimized for answering questions about heart failure. It has been trained to provide medically relevant information about heart failure symptoms, causes, treatments, and related cardiovascular concepts.

Intended Use

Primary Use Cases

• Educational Q&A about heart failure
• Medical information assistance for healthcare professionals
• Patient education support (under professional supervision)

Out-of-Scope Use Cases

• Clinical diagnosis or treatment decisions
• Emergency medical situations
• Replacement for professional medical advice

Performance Metrics

Core Language Metrics

• ROUGE-1: 0.3003
• ROUGE-2: 0.1253
• ROUGE-L: 0.2335
• BLEU: 0.0681
• Semantic Similarity: 0.4325
• Perplexity: 6.49

Domain-Specific Metrics

• Medical Term Usage: 100.0% (cardiac domain relevance)
• Response Completeness: 100.0% (proper sentence endings)
• Average Response Length: 45.1 words
• Repetition Score: 1.14 (lower is better)

Overall Performance Score: 0.182

Training Data

• Evaluation Samples: 20 heart failure Q&A pairs
• Domain Focus: Cardiovascular health, specifically heart failure
• Data Quality: Medically accurate reference answers

Model Architecture

• Tokenizer Vocab Size: 32,001
• Custom Tokens: <|end|> token for response termination
• Device: CPU optimized
• Generation Parameters:
  • Temperature: 0.6
  • Top-p: 0.9
  • Max tokens: 80