yongxianwei/Qwen2.5-VL-32B-VQA
yongxianwei/Qwen2.5-VL-32B-VQA is a 32-billion parameter vision-language model, specifically fine-tuned as a Visual Question Answering (VQA) expert. Derived from the Qwen2.5-VL-32B-Instruct base model, it is part of the OptMerge MLLM merging benchmark, designed to be merged with other specialized experts without additional training data. This model excels in VQA tasks, demonstrating improved performance on datasets like GQA and TextVQA compared to its base model.
Loading preview...
Overview
yongxianwei/Qwen2.5-VL-32B-VQA is a 32-billion parameter multimodal large language model (MLLM) specialized in Visual Question Answering (VQA). It is an expert model developed by yongxianwei as part of the OptMerge MLLM merging benchmark, which explores unifying diverse multimodal capabilities through model merging. This model is fine-tuned from the Qwen/Qwen2.5-VL-32B-Instruct base model.
Key Capabilities
- Specialized VQA Performance: Optimized for visual question answering tasks, showing enhanced performance on benchmarks such as GQA and TextVQA.
- OptMerge Expert: Designed to be merged with other specialized experts (Geometry, Chart, OCR, Grounding) using the OptMerge framework, allowing for a unified multimodal model without requiring new training data.
- Full Weights: Provided as full weights, eliminating the need for PEFT adapters for direct loading and use.
Good for
- Visual Question Answering Applications: Ideal for scenarios requiring precise answers to questions based on visual input.
- Research on Model Merging: A valuable component for researchers exploring the OptMerge framework and the unification of MLLM capabilities.
- Building Composite Multimodal Systems: Can be integrated into larger systems that combine various multimodal experts for comprehensive understanding.