Spaces:
Running on Zero
Running on Zero
| title: EVisRAG Multi-Image VQA | |
| emoji: 🔍 | |
| colorFrom: blue | |
| colorTo: pink | |
| sdk: gradio | |
| sdk_version: 6.21.0 | |
| app_file: app.py | |
| short_description: Evidence-guided multi-image visual question answering | |
| python_version: "3.12" | |
| startup_duration_timeout: 30m | |
| # EVisRAG-7B: Evidence-Guided Multi-Image VQA | |
| This Space runs **EVisRAG-7B** (VisRAG 2.0), a multi-image visual question-answering model that | |
| mitigates visual hallucinations via evidence-guided reasoning. | |
| Given a question and 1–5 retrieved images, the model: | |
| 1. **Observes** each image for relevant information | |
| 2. **Extracts evidence** from each image | |
| 3. **Reasons** over the collected evidence | |
| 4. **Answers** the question grounded in the evidence | |
| Based on [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), fine-tuned | |
| with Reward-Scoped GRPO. | |
| - Paper: [VisRAG2.0](https://huggingface.co/papers/2510.09733) | |
| - Model: [openbmb/EVisRAG-7B](https://huggingface.co/openbmb/EVisRAG-7B) | |
| - Code: [GitHub](https://github.com/OpenBMB/VisRAG) |