wangyueqian/MMDuet2
MMDuet2 by wangyueqian is a 3 billion parameter Video Multimodal Large Language Model (MLLM) designed for proactive interaction, allowing it to respond not only after user input but also at appropriate moments during video playback. Post-trained from Qwen2.5-VL-3B-Instruct, this lightweight model offers fast, real-time interaction, addressing issues of sparse or overly dense responses common in previous video MLLMs. Its primary use case is enabling dynamic and timely conversational engagement with video content.
Loading preview...
MMDuet2: Proactive Video Interaction
MMDuet2 is a 3 billion parameter Video Multimodal Large Language Model (MLLM) developed by wangyueqian, specifically engineered for proactive interaction with video content. Unlike traditional models that only respond after a user's turn, MMDuet2 can generate timely and appropriate replies at any point during video playback, enhancing the conversational experience.
Key Capabilities
- Proactive Interaction: Responds dynamically during video playback, not just after user input.
- Lightweight & Fast: With only 3B parameters, it's designed for real-time interaction.
- Balanced Responses: Addresses common issues of overly sparse or dense/repetitive replies found in other video MLLMs.
- Video MLLM: Post-trained from Qwen2.5-VL-3B-Instruct to specialize in video understanding and generation.
Good for
- Applications requiring dynamic, real-time conversational agents interacting with video streams.
- Scenarios where timely, context-aware responses during video playback are crucial.
- Developers seeking a lightweight yet capable video MLLM for interactive systems.