Science Question Answering on SQA-I
79ScorePreSel
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PreSelModel=Llama-3-8B, Sampling Ratio=15%2025.03 | 79 | — | |
| RandomModel=Llama-3-8B, Sampling Ratio=15%2025.03 | 78.3 | — | |
| Full FinetuneModel=Llama-3-8B, Sampling Ratio=100%2025.03 | 78.2 | — | |
| VisNecBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 74.5 | — | |
| MAGICTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 73.5 | — | |
| LLaVA-1.5-VL-13BBackbone=LLaVA-1.5-VL-13B, Token=5762026.03 | 72.88 | 100 | |
| Full FinetuneModel=Vicuna-13B, Sampling Ratio=100%2025.03 | 72 | — | |
| TPRLBackbone=LLaVA-1.5-VL-13B, Token=192*2026.03 | 71.8 | 98.5 | |
| Full-FinetuneTarget Model=LLaVA-1.5-13B, Data Budget=100%2026.05 | 71.2 | — | |
| PreSelModel=Vicuna-13B, Sampling Ratio=15%2025.03 | 70.8 | — | |
| TPRLBackbone=LLaVA-1.5-VL-13B, Token=128*2026.03 | 70.8 | 97.1 | |
| SemDedupTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 70.5 | — | |
| D2-PruningTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 70.4 | — | |
| Self-SupTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 70.2 | — | |
| COINCIDETarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 70 | — | |
| RandomModel=Vicuna-13B, Sampling Ratio=15%2025.03 | 69.9 | — | |
| FastVBackbone=LLaVA-1.5-VL-13B, Token=1922026.03 | 69.9 | 92.1 | |
| EL2NTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 69.9 | — | |
| CLIP-ScoreTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 69.7 | — | |
| Self-FilterTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 69.5 | — | |
| RandomTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 68.8 | — | |
| PerplexityTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.05 | 68.7 | — | |
| FastVBackbone=LLaVA-1.5-VL-13B, Token=1282026.03 | 68 | 93.3 | |
| PreSelBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 66.2 | — | |
| CoIDOBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 66.2 | — | |
| COINCIDEBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 63.9 | — | |
| EL2NBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 62.7 | — | |
| CLIP-ScoreBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 62.3 | — | |
| RandomBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 62.1 | — | |
| PRISMSample Count=57K2025.02 | 61.1 | — | |
| Full DataSample Count=186K2025.02 | 60.8 | — | |
| ICONSBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 60.2 | — | |
| Self-FilterBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 59.3 | — | |
| TypiClustBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 59.2 | — | |
| XMASBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 58.1 | — | |
| IFDBackbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=15% (28K)2026.03 | 57.8 | — | |
| RandomSample Count=57K2025.02 | 56.5 | — | |
| Full-Data(186K)Backbone=LLaVA-v1.5-7B, Training Dataset=Vision-Flan-186K, Sampling Ratio=100%2026.03 | 55.6 | — |