longvideoagent/longvideoagent-qwen2.5-7b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Mar 22, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The longvideoagent/longvideoagent-qwen2.5-7b is a 7.6 billion parameter Qwen2.5-7B-based language model checkpoint specifically fine-tuned for long-video question answering within the LongVideoAgent multi-agent framework. Developed by the LongVideoAgent project team, this model excels at reasoning over extended video content by integrating with specialized agents for planning, temporal grounding, and visual evidence extraction. It demonstrates a 6.66 percentage point improvement in LongTVQA+ accuracy over the Qwen2.5-7B-Instruct baseline, making it suitable for research and reproduction of long-video QA experiments.

Loading preview...

LongVideoAgent Qwen2.5-7B Checkpoint

This repository provides the specific Qwen2.5-7B-based language model checkpoint utilized within the LongVideoAgent multi-agent framework. LongVideoAgent is designed for complex question answering over long video content, decomposing reasoning into specialized roles.

Key Capabilities & Architecture

  • Specialized for Long-Video QA: This 7.6 billion parameter model is fine-tuned to support the LongVideoAgent's approach to understanding and answering questions about extended videos.
  • Multi-Agent Integration: It functions as a core component within a framework that includes a MasterAgent for planning, a GroundingAgent for subtitle-based temporal grounding, and a VisionAgent for extracting local visual evidence.
  • Performance Improvement: The model demonstrates a notable performance gain, improving LongTVQA+ accuracy by 6.66 percentage points compared to the base Qwen2.5-7B-Instruct model, achieving 64.00% accuracy.

Intended Use Cases

  • Research: Ideal for academic research into long-video question answering and multi-agent reasoning systems.
  • Reproducibility: Designed for reproducing experiments conducted with the LongVideoAgent framework.
  • Agentic Reasoning Studies: Useful for studying how agentic approaches can be applied to complex video understanding tasks.

Note: This checkpoint is not a standalone general-purpose video model. For full functionality, it requires integration with the official LongVideoAgent codebase and evaluation pipeline, available on their project page and GitHub repository.

Finetunes
1 models