List your company

Text-based Visual Question Answering on TextVQA (Score & Avg. Rel.)

85.2ScoreFull Model
45.36855.70966.0576.391Feb 17, 2025May 9, 2025Jul 30, 2025Oct 20, 2025Jan 10, 2026Apr 2, 2026Jun 23, 2026

Evaluation results

MethodLinks
Full ModelRatio=1.02026.0485.2100
HAWKRatio=0.62026.048599.1
HAWKRatio=0.82026.048395.8
VanillaToken Reduction Ratio=Full Tokens (100%), Model=Qwen2-VL-72B, Quantization=4-bit2026.0682—
HAWKRatio=0.92026.0479.888.5
FastVToken Reduction Ratio=↓ 66.7%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0679.1—
PDropToken Reduction Ratio=↓ 66.7%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0678.9—
CLSEToken Reduction Ratio=↓ 66.7%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0678.5—
PDropToken Reduction Ratio=↓ 77.8%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0676.7—
FastVToken Reduction Ratio=↓ 77.8%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0676.6—
VanillaBackbone=InternVL-2.5-8B, Token Reduction Rate=100%2026.0476.23—
CLSEToken Reduction Ratio=↓ 77.8%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0676—
VisionZipRatio=0.62026.0475.396.1
FastVBackbone=InternVL-2.5-8B, Token Reduction Rate=25%2026.0474.65—
DARTRatio=0.62026.0473.889.5
DASHBackbone=InternVL-2.5-8B, Token Reduction Rate=25%2026.0473.28—
CLSEToken Reduction Ratio=↓ 88.9%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0672.1—
DivPruneBackbone=InternVL-2.5-8B, Token Reduction Rate=25%2026.0470.43—
VisionZipRatio=0.82026.0470.189.2
PDropToken Reduction Ratio=↓ 88.9%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0670—
FastVToken Reduction Ratio=↓ 88.9%, Model=Qwen2-VL-72B, Quantization=4-bit2026.0668.8—
DARTRatio=0.82026.0467.280.4
DARTBackbone=InternVL-2.5-8B, Token Reduction Rate=25%2026.0464.74—
VisionZipRatio=0.92026.0461.477.5
VanillaTokens=2880, Retaining Ratio=100%, Backbone=LLaVA-NeXT2025.1161.3100
OC-VTPTokens=640, Retaining Ratio=22.2%, Individual Training=true, Backbone=LLaVA-NeXT2025.1161.199.7
VanillaModel=LLaVA-Next-7B, Token Budget=2880, Reduction Ratio=0%2026.0560.3100
VisionZipTokens=640, Retaining Ratio=22.2%, Backbone=LLaVA-NeXT2025.1160.298.2
OC-VTPTokens=640, Retaining Ratio=22.2%, Backbone=LLaVA-NeXT2025.1160.198
HiPruneTokens=640, Retaining Ratio=22.2%, Backbone=LLaVA-NeXT2025.116097.9
DARTRatio=0.92026.0459.872.2
SPprunerModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.0559.397.7
VisionZipTokens=320, Retaining Ratio=11.1%, Backbone=LLaVA-NeXT2025.1158.996.1
VisionZipModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.0558.896.2
DARTModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.0558.496.4
OC-VTPTokens=320, Retaining Ratio=11.1%, Individual Training=true, Backbone=LLaVA-NeXT2025.1158.495.3
VanillaToken Budget (Retained Tokens)=576, Reduction Ratio (↓ %)=0%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0558.2100
OC-VTPTokens=320, Retaining Ratio=11.1%, Backbone=LLaVA-NeXT2025.1158.294.9
HiPruneTokens=320, Retaining Ratio=11.1%, Backbone=LLaVA-NeXT2025.1157.694
SPprunerToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0557.599.4
DARTToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0557.399
VisionZipToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0557.299
PyramidDropToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0556.998.7
SPprunerToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0556.998.5
PyramidDropModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.0556.990.1
VisionZipToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0556.897.6
OC-VTPTokens=160, Retaining Ratio=5.6%, Individual Training=true, Backbone=LLaVA-NeXT2025.1156.892.7
PyramidDropToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0556.796.5
DivPruneToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0556.698.8
DARTToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0556.697.2
DivPruneModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.0556.296.1
DivPruneToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.055697.4
FastVModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.0555.786.8
OC-VTPTokens=160, Retaining Ratio=5.6%, Backbone=LLaVA-NeXT2025.1155.790.9
VisionZipToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0555.690.4
PACTToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0555.598.1
VisionZipTokens=160, Retaining Ratio=5.6%, Backbone=LLaVA-NeXT2025.1155.590.5
SPprunerToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0555.495.2
PACTToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0554.996.9
DivPruneToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0554.694.4
DARTToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0554.193.4
PyramidDropToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0553.788.6
PACTToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0553.294
FastVToken Budget (Retained Tokens)=192, Reduction Ratio (↓ %)=66.7%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0552.590
HiPruneTokens=160, Retaining Ratio=5.6%, Backbone=LLaVA-NeXT2025.1151.283.5
PRISMSample Count=57K2025.0250.8—
FastVToken Budget (Retained Tokens)=128, Reduction Ratio (↓ %)=77.8%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0550.682.9
Full DataSample Count=186K2025.0250.4—
FastVToken Budget (Retained Tokens)=64, Reduction Ratio (↓ %)=88.9%, Base Model (LLaVA-1.5-7B)=LLaVA-1.5-7B2026.0547.874
RandomSample Count=57K2025.0246.9—