Text-Rich VQA on AI2D
71.9ScoreScaled base VLM 7B + MERIT (Proposed, 2D)
Evaluation results
| Method | Links | |
|---|---|---|
| Scaled base VLM 7B + MERIT (Proposed, 2D)Train Data=3.6M + 1.6M2026.06 | 71.9 | |
| Scaled base VLM 7B + Joint FFTTrain Data=3.6M + 1.6M2026.06 | 71.6 | |
| Scaled base VLM 7BTrain Data=3.6M2026.06 | 69.2 | |
| LLaVA-1.5-13BTrain Data=0.7M2026.06 | 59.5 | |
| Base VLM 7B + Joint FFTTrain Data=0.7M + 1.6M2026.06 | 58.6 | |
| Base VLM 7B + MERIT (Proposed, 2D)Train Data=0.7M + 1.6M2026.06 | 58.4 | |
| LLaVA-1.5-7BTrain Data=0.7M2026.06 | 54.8 | |
| LLaVA-7BTrain Data=0.6M2026.06 | 48.3 | |
| Base VLM 7BTrain Data=0.7M2026.06 | 46.5 |