# Mixed Preference Optimization

InternVL2.5-MPO is a series of multimodal large language models based on InternVL2.5 and Mixed Preference Optimization (MPO). It excels in multimodal tasks by integrating the recently incrementally pre-trained InternViT with various pre-trained large language models (LLMs) such as InternLM 2.5 and Qwen 2.5, utilizing a randomly initialized MLP projector. This model series has been trained on the multimodal reasoning preference dataset MMPR, which contains approximately 3 million samples, enhancing the model's reasoning capabilities and answer quality through an effective data construction process and mixed preference optimization techniques.

InternVL2_5-26B-MPO is a multimodal large language model (MLLM) that builds upon InternVL2.5 and improves model performance through Mixed Preference Optimization (MPO). The model can handle multimodal data, including images and text, and is widely applied in scenarios such as image captioning and visual question answering. Its significance lies in its ability to understand and generate text closely related to image content, pushing the boundaries of multimodal AI. Background information on the product includes its exceptional performance in multimodal tasks and evaluation results on the OpenCompass Leaderboard. This model provides researchers and developers with a powerful tool to explore and realize the potential of multimodal AI.

The InternVL2_5-8B-MPO-AWQ is a multimodal large language model launched by OpenGVLab, based on the InternVL2.5 series and utilizing Mixed Preference Optimization (MPO) technology. This model demonstrates exceptional performance in understanding and generating both visual and language content, particularly excelling in multimodal tasks. It combines the visual component InternViT with the linguistic component InternLM or Qwen, employing a randomly initialized MLP projector for incremental pre-training, enabling in-depth understanding and interaction with images and texts. The significance of this technology lies in its capacity to handle various data types, including single images, multiple images, and video data, providing new solutions for the multimodal AI field.