Visual Question Answering on ScienceVQA (test)
85.9AccuracyGRIP (Idefics2-8B)
Evaluation Results
| Method | Links | |
|---|---|---|
| GRIP (Idefics2-8B)Target Model=GPT 4o, Retriever Training Source=Idefics2-8B2026.06 | 85.9 | |
| GRIP (Qwen2.5-VL-7B)Target Model=GPT 4o, Retriever Training Source=Qwen2.5-VL-7B2026.06 | 85 | |
| CLIP Similarity BasedTarget Model=GPT 4o, Retrieval Strategy=CLIP Similarity2026.06 | 84.8 | |
| Zero ShotTarget Model=GPT 4o, Retrieval Strategy=Zero Shot2026.06 | 79.3 | |
| GRIP (Qwen2.5-VL-7B)Target Model=Gemini 2.5 Flash Lite, Retriever Training Source=Qwen2.5-VL-7B2026.06 | 75.9 | |
| GRIP (Idefics2-8B)Target Model=Gemini 2.5 Flash Lite, Retriever Training Source=Idefics2-8B2026.06 | 75.2 | |
| CLIP Similarity BasedTarget Model=Gemini 2.5 Flash Lite, Retrieval Strategy=CLIP Similarity2026.06 | 75.1 | |
| Zero ShotTarget Model=Gemini 2.5 Flash Lite, Retrieval Strategy=Zero Shot2026.06 | 67.1 |