promotion/Llama-3.1-8B-TLDR-AbsoluteMaxMin-baseline
The promotion/Llama-3.1-8B-TLDR-AbsoluteMaxMin-baseline model is an 8 billion parameter language model based on the Llama-3.1-8B-Instruct architecture, fine-tuned for TL;DR summarization tasks. It focuses on optimizing objectives like coverage, faithfulness, conciseness, and helpfulness. This model is specifically designed for generating high-quality, concise summaries, differing from other models in its objective weighting strategy during optimization.
Loading preview...
Model Overview
This model, promotion/Llama-3.1-8B-TLDR-AbsoluteMaxMin-baseline, is an 8 billion parameter language model built upon the meta-llama/Llama-3.1-8B-Instruct backbone. It is specifically fine-tuned for TL;DR (Too Long; Didn't Read) summarization tasks, aiming to produce concise, faithful, and helpful summaries.
Key Characteristics
- Objective-driven Optimization: The model's training focuses on optimizing specific objectives: coverage, faithfulness, conciseness, and helpfulness, using an AbsoluteMaxMin strategy for objective weighting.
- Context Length: Supports a context length of 32,768 tokens.
- Training Details: Underwent 300 optimizer updates with a global batch size of 16, using the same response pool, temperatures, and disagreement point as NBPO (Nash Bargaining Preference Optimization).
Evaluation
Evaluated for independent objective-wise win rate against a common reference, with judgments provided by Llama-3.3-70B-Instruct on prompt-disjoint held-out prompts. This model's performance is reported in the Nash Bargaining Preference Optimization (NBPO) paper, specifically in Table 2 for cross-method evaluation.
Ideal Use Cases
This model is particularly well-suited for applications requiring:
- High-quality summarization: Generating accurate and concise TL;DR summaries from longer texts.
- Objective-focused text generation: Where specific metrics like faithfulness and conciseness are paramount.