Text-based Visual Question Answering on TextVQA (VQAText & Average Metrics)
82.7VQAText ScoreFull Precision
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Full PrecisionModel=InternVL3.5-38B-Instruct, Weight bits=16 bit, Memory=71.51 GB2026.07 | 82.7 | — | |
| GPTQModel=InternVL3.5-38B-Instruct, Weight bits=3 bit, Memory=12.71 GB2026.07 | 80.22 | — | |
| SAB-LVLMModel=InternVL3.5-38B-Instruct, Weight bits=1.06 bit, Memory=10.27 GB2026.07 | 78.06 | — | |
| Full PrecisionModel=InternVL3.5-14B-Instruct, Weight bits=16 bit, Memory=28.16 GB2026.07 | 77.8 | — | |
| Full PrecisionModel=InternVL3.5-8B-Instruct, Weight bits=16 bit, Memory=15.89 GB2026.07 | 77.38 | — | |
| ARB-LLMModel=InternVL3.5-38B-Instruct, Weight bits=1.08 bit, Memory=10.27 GB2026.07 | 77.37 | — | |
| GPTQModel=InternVL3.5-14B-Instruct, Weight bits=3 bit, Memory=5.38 GB2026.07 | 74.9 | — | |
| GPTQModel=InternVL3.5-8B-Instruct, Weight bits=3 bit, Memory=2.83 GB2026.07 | 74.82 | — | |
| SAB-LVLMModel=InternVL3.5-14B-Instruct, Weight bits=1.07 bit, Memory=4.33 GB2026.07 | 71.03 | — | |
| ARB-LLMModel=InternVL3.5-14B-Instruct, Weight bits=1.07 bit, Memory=4.33 GB2026.07 | 70.77 | — | |
| PB-LLMModel=InternVL3.5-38B-Instruct, Weight bits=1.70 bit, Memory=10.24 GB2026.07 | 65.06 | — | |
| LLaVA-Next-7BBase Model=LLaVA-Next-7B, Reduction Ratio=100%, Retained Tokens=28802026.06 | 64.9 | 100 | |
| VisPrunerBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 59.1 | 90.8 | |
| HoloVBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 57.1 | 87.5 | |
| VisPrunerBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 57.1 | 86.7 | |
| HoloV + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 56.1 | 92 | |
| HoloVBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 56 | 86.3 | |
| VisPruner + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 55.9 | 91.9 | |
| VisionzipBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 55.2 | 88.9 | |
| VisionzipBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 54.5 | 84.9 | |
| HoloV + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 53.1 | 89 | |
| Visionzip + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 53 | 90.5 | |
| VisPruner + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 52 | 88.5 | |
| DivPruneBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 51.6 | 88.8 | |
| DivPrune + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=11.1%, Retained Tokens=3202026.06 | 51.4 | 90.2 | |
| Visionzip + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 51.2 | 87.9 | |
| DivPruneBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 51.2 | 86.3 | |
| DivPrune + RESTOREBase Model=LLaVA-Next-7B, Reduction Ratio=5.5%, Retained Tokens=1602026.06 | 48.3 | 87.2 | |
| SAB-LVLMModel=InternVL3.5-8B-Instruct, Weight bits=1.07 bit, Memory=2.28 GB2026.07 | 42.04 | — | |
| BiLLMModel=InternVL3.5-38B-Instruct, Weight bits=1.07 bit, Memory=10.72 GB2026.07 | 37.09 | — | |
| BiLLMModel=InternVL3.5-14B-Instruct, Weight bits=1.07 bit, Memory=4.53 GB2026.07 | 31.65 | — | |
| ARB-LLMModel=InternVL3.5-8B-Instruct, Weight bits=1.07 bit, Memory=2.28 GB2026.07 | 30.83 | — | |
| BiLLMModel=InternVL3.5-8B-Instruct, Weight bits=1.08 bit, Memory=2.39 GB2026.07 | 0.23 | — | |
| PB-LLMModel=InternVL3.5-8B-Instruct, Weight bits=1.70 bit, Memory=2.28 GB2026.07 | 0 | — | |
| PB-LLMModel=InternVL3.5-14B-Instruct, Weight bits=1.70 bit, Memory=4.34 GB2026.07 | 0 | — |