csankalp21/headlinegpt
csankalp21/headlinegpt is a 1.5 billion parameter language model, fine-tuned from Qwen/Qwen2.5-1.5B-Instruct, specifically designed for generating concise and engaging titles. It utilizes reward-weighted supervised fine-tuning on content-title pairs to optimize for high-engagement headline characteristics. This model excels at creating attention-grabbing titles for news articles, academic content, presentations, and social media posts.
Loading preview...
HeadlineGPT Overview
HeadlineGPT is a specialized 1.5 billion parameter language model, fine-tuned from the Qwen2.5-1.5B-Instruct base model. Its primary function is to generate concise and engaging titles from provided source content.
Key Capabilities
- High-Engagement Title Generation: The model was trained using a unique reward-weighted supervised fine-tuning approach. This method prioritizes training examples with higher engagement scores, encouraging the generation of titles that are more likely to capture attention.
- Specialized Fine-tuning: Fine-tuned using LoRA (Low-Rank Adaptation) with specific parameters (rank 16, alpha 32, dropout 0.05) on a dataset of content-title pairs.
- Efficient Training: Trained for 2 epochs on a 25,000-example subset, focusing on English language content.
Good For
HeadlineGPT is particularly well-suited for applications requiring compelling short-form titles across various domains:
- News and Article Headlines: Crafting catchy titles for journalistic content.
- Academic and Research Titles: Generating clear and engaging titles for papers, talks, and presentations.
- Social Media Posts: Creating attention-grabbing captions or titles for platforms like Twitter, LinkedIn, or blogs.
- General Content Titling: Any scenario where a concise, high-engagement title is needed for longer-form content.