Recently, research teams from Peking University announced the release of an open-source multimodal model called LLaVA-o1, which is claimed to be the first visual language model capable of spontaneous and systematic reasoning, comparable to GPT-o1. The model excels in six challenging multimodal benchmark tests, with its 11B parameter version outperforming competitors such as Gemini-1.5-pro, GPT-4o-mini, and Llama-3.2-90B-Vision-Instruct.

PaddleMIX 2.0 is a multi-modal model development kit launched by Baidu, aimed at simplifying multi-modal application development and supporting scenarios such as autonomous driving, smart healthcare, and search engines. Key highlights include: 1. **Rich Model Library**: Covers images, text, video, and audio, with the addition of LLaVA series models, providing cutting-edge technology support. 2. **Full Process Development Experience**: Equipped with DataCopilot and Auto modules to simplify the training process of multi-modal models. 3. **High Performance Training Capability**

Microsoft has open-sourced the multimodal model LLaVA-1.5, inheriting the LLaVA architecture and introducing new features. Researchers have tested it in visual question answering, natural language processing, image generation, and other areas, showing that LLaVA-1.5 has reached the highest level among open-source models.

💡 The development trend of open source large models: Open source models like LLaVA 1.5 are rapidly advancing and are expected to become strong competitors to commercial closed-source models like GPT-4 Vision. 🔥 Technical highlights of LLaVA 1.5: it integrates the CLIP image encoder, LLaMA language model, and MLP connector, achieving impressive multimodal benchmark results with just 600k training samples in one day. 📈 Efficiency advantages of LLaVA 1.5: compared to other open source large models, it has higher computational efficiency and lower resource consumption.

["Zhejiang University alumni collaborated with Microsoft Research to launch the multimodal model LLaVA, challenging GPT-4V.","LLaVA has demonstrated excellent performance on 11 testing datasets, garnering over 6k stars.","The model exhibits comprehensive capabilities, exceeding 85% of the level of GPT-4V.","The open-source code, model, and training data for LLaVA are now available for use."]