promotion/Llama-3.1-8B-TLDR-NBPO-600updates

TEXT GENERATIONPricing:Input $0.2 / Cached $0.028 / Output $0.32Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 7, 2026License:llama3.1Architecture:Transformer Featherless Exclusive Cold

This model is an 8 billion parameter Llama-3.1-8B-Instruct backbone fine-tuned for Nash Bargaining Preference Optimization (NBPO) on TL;DR tasks. It focuses on optimizing for coverage, faithfulness, conciseness, and helpfulness, demonstrating improved held-out fit compared to its reference policy. The model is specifically designed for summarization tasks, aiming to balance multiple objective-wise win rates.

Loading preview...

Model Overview

This model, based on the meta-llama/Llama-3.1-8B-Instruct backbone, is an 8 billion parameter language model fine-tuned using Nash Bargaining Preference Optimization (NBPO) specifically for TL;DR (Too Long; Didn't Read) summarization tasks. It underwent 300 optimizer updates with a global batch size of 16.

Key Capabilities

  • Multi-objective Optimization: The model is optimized across four key objectives: coverage, faithfulness, conciseness, and helpfulness, aiming for a balanced performance across these metrics.
  • Improved Held-out Fit: Demonstrates an improved held-out fit (nMSE 0.8960 vs 0.9022) compared to its reference policy, indicating better generalization on unseen data.
  • Robust Evaluation: Evaluated using an independent objective-wise win rate against the common reference, judged by Llama-3.3-70B-Instruct on prompt-disjoint held-out prompts, considering both presentation orders.

Good For

  • Summarization Tasks: Particularly well-suited for generating concise and informative summaries, balancing aspects like content coverage and factual accuracy.
  • Research on Preference Optimization: Useful for researchers exploring the effects of Nash Bargaining Preference Optimization on multi-objective language model fine-tuning, especially concerning budget sensitivity.