SummLlama3.1-8B
| Property | Value |
|---|---|
| Parameter Count | 8.03B |
| Base Model | Llama-3.1-8B-Instruct |
| Training Method | Direct Preference Optimization (DPO) |
| Paper | Research Paper |
| Tensor Type | BF16 |
What is SummLlama3.1-8B?
SummLlama3.1-8B is a specialized language model fine-tuned for generating high-quality text summaries. Built upon Llama-3.1-8B-Instruct, it has been optimized using Direct Preference Optimization with over 100K summarization feedback samples across seven distinct domains including news, lifestyle, medical, and various dialogue formats.
Implementation Details
The model leverages LLM-generated feedback instead of expensive human annotations, focusing on three critical aspects: faithfulness (0.924), completeness (0.635), and conciseness (0.661). Automated evaluations show significant improvements over the base model, achieving a combined score of 0.740.
- Specialized prompt template for optimal performance
- Multi-domain training across 7 different content types
- Optimized for both dialogue and non-dialogue summarization
Core Capabilities
- Generates human-preferred summaries with high faithfulness
- Maintains key information while remaining concise
- Performs well across multiple content domains
- Supports both short and long-form text summarization
Frequently Asked Questions
Q: What makes this model unique?
The model's unique strength lies in its optimization through DPO using large-scale LLM-generated feedback, resulting in summaries that outperform both Llama3-8B and Llama3-70B in automated evaluations.
Q: What are the recommended use cases?
The model is ideal for summarizing various content types including news articles, medical texts, lifestyle content, and dialogues (meetings, interviews, daily conversations). It's particularly effective when faithful and concise summaries are required.
