List your company

Scientific Question Answering on ScienceQA

94.4AccuracyVanilla
-3.77621.71247.272.688May 30, 2024Oct 3, 2024Feb 7, 2025Jun 14, 2025Oct 19, 2025Feb 23, 2026Jun 30, 2026

Evaluation results

MethodLinks
VanillaBackbone=Qwen3-VL-8B, TFLOPs=100%2026.0694.4
Op-SkipBackbone=Qwen3-VL-8B, TFLOPs=66%, N=202026.0694.35
ShortVBackbone=Qwen3-VL-8B, TFLOPs=58%, N=202026.0692.61
APETBackbone=Qwen3-VL-8B, TFLOPs=66%2026.0692.56
VSkipBackbone=Qwen3-VL-8B, TFLOPs=76%, N=202026.0692.46
V2DropBackbone=Qwen3-VL-8B, TFLOPs=66%2026.0691.72
MASQuantModel=Qwen2.5-VL-7B, Bits=W8A82026.0388.6
MBQModel=Qwen2.5-VL-7B, Bits=W8A82026.0388.5
DenseModel=Qwen2.5-VL-7B, Bits=FP162026.0388.4
SQModel=Qwen2.5-VL-7B, Bits=W8A82026.0388.2
RTNModel=Qwen2.5-VL-7B, Bits=W8A82026.0388.1
AWQModel=Qwen2.5-VL-7B, Bits=W4A162026.0387.8
MBQModel=Qwen2.5-VL-7B, Bits=W4A162026.0387.8
MASQuantModel=Qwen2.5-VL-7B, Bits=W4A162026.0387.8
VanillaBackbone=Qwen2.5-VL-7B, TFLOPs=100%2026.0687.51
APETBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.0687.51
V2DropBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.0687.06
MBQModel=Qwen2.5-VL-7B, Bits=W4A82026.0386
VSkipBackbone=Qwen2.5-VL-7B, TFLOPs=67%, N=162026.0685.72
MASQuantModel=Qwen2.5-VL-7B, Bits=W4A82026.0385.7
Op-SkipBackbone=Qwen2.5-VL-7B, TFLOPs=55%, N=202026.0685.52
RTNModel=Qwen2.5-VL-7B, Bits=W4A82026.0385.2
SQModel=Qwen2.5-VL-7B, Bits=W4A82026.0383.3
MASQuantModel=Qwen2.5-VL-3B, Bits=W4A162026.0382.4
RTNModel=Qwen2.5-VL-3B, Bits=W8A82026.0382.4
MASQuantModel=Qwen2.5-VL-3B, Bits=W8A82026.0382.4
SQModel=Qwen2.5-VL-3B, Bits=W8A82026.0382.2
DenseModel=Qwen2.5-VL-3B, Bits=FP162026.0381.9
MBQModel=Qwen2.5-VL-3B, Bits=W8A82026.0381.8
AWQModel=Qwen2.5-VL-3B, Bits=W4A162026.0381.7
MBQModel=Qwen2.5-VL-3B, Bits=W4A162026.0381.5
RTNModel=Qwen2.5-VL-3B, Bits=W4A162026.0381.3
RTNModel=Qwen2.5-VL-7B, Bits=W4A162026.0381.3
VTWBackbone=Qwen3-VL-8B, TFLOPs=58%, K=162026.0681.11
VTWBackbone=Qwen2.5-VL-7B, TFLOPs=50%, K=162026.0680.32
MBQModel=Qwen2.5-VL-7B, Bits=W4A62026.0380.1
MASQuantModel=Qwen2.5-VL-7B, Bits=W4A62026.0379.7
SQModel=Qwen2.5-VL-7B, Bits=W4A62026.0378.6
ShortVBackbone=Qwen2.5-VL-7B, TFLOPs=50%, N=202026.0678.38
MASQuantModel=Qwen2.5-VL-3B, Bits=W4A82026.0377.9
MBQModel=Qwen2.5-VL-3B, Bits=W4A82026.0376.7
LLaVA-v1.6 (7B) w/ STICBase Model=LLaVA-v1.6 (7B), Strategy=STIC2024.0575.3
SQModel=Qwen2.5-VL-3B, Bits=W4A82026.0372.1
MASQuantModel=Qwen2.5-VL-3B, Bits=W4A62026.0371.2
PDropToken Retention Rate=64, Base Model=LLaVA-1.5-7B2025.0669.96
SQModel=Qwen2.5-VL-3B, Bits=W4A62026.0369.9
VTWBackbone=LLaVA-1.5-7B, TFLOPs=55%, K=162026.0669.71
SparseVLM-v1.5Token Retention Rate=64, Base Model=LLaVA-1.5-7B2025.0669.61
PDropToken Retention Rate=192, Base Model=LLaVA-1.5-7B2025.0669.56
LLaVA-1.5-7BToken Retention Rate=576, Base Model=LLaVA-1.5-7B2025.0669.51
LLaVA-v1.5 (7B) w/ STICBase Model=LLaVA-v1.5 (7B), Strategy=STIC2024.0569.5
VanillaBackbone=LLaVA-1.5-7B, TFLOPs=100%2026.0669.46
VisionDropToken Retention Rate=32, Base Model=LLaVA-1.5-7B2025.0669.41
PDropToken Retention Rate=128, Base Model=LLaVA-1.5-7B2025.0669.36
FastVToken Retention Rate=64, Base Model=LLaVA-1.5-7B2025.0669.36
VisionDropToken Retention Rate=64, Base Model=LLaVA-1.5-7B2025.0669.31
VisPrunerToken Retention Rate=32, Base Model=LLaVA-1.5-7B2025.0669.26
VSkip+Backbone=LLaVA-1.5-7B, TFLOPs=76%, N=202026.0669.26
VSkipBackbone=LLaVA-1.5-7B, TFLOPs=76%, N=202026.0669.16
Op-SkipBackbone=LLaVA-1.5-7B, TFLOPs=66%, N=202026.0669.16
V2DropBackbone=LLaVA-1.5-7B, TFLOPs=66%2026.0669.12
VisionDropToken Retention Rate=192, Base Model=LLaVA-1.5-7B2025.0669.06
VisPrunerToken Retention Rate=128, Base Model=LLaVA-1.5-7B2025.0669.01
VisionZipToken Retention Rate=64, Base Model=LLaVA-1.5-7B2025.0669.01
APETBackbone=LLaVA-1.5-7B, TFLOPs=66%2026.0668.93
LLaVA-v1.6 (7B)Model=LLaVA-v1.6 (7B)2024.0568.9
SparseVLM-v1.5Token Retention Rate=192, Base Model=LLaVA-1.5-7B2025.0668.86
VisionZipToken Retention Rate=192, Base Model=LLaVA-1.5-7B2025.0668.86
VisionZipToken Retention Rate=128, Base Model=LLaVA-1.5-7B2025.0668.82
VisPrunerToken Retention Rate=64, Base Model=LLaVA-1.5-7B2025.0668.82
LLaVA-v1.5 (7B) w/ POVIDBase Model=LLaVA-v1.5 (7B), Strategy=POVID2024.0568.8
VisionZipToken Retention Rate=32, Base Model=LLaVA-1.5-7B2025.0668.72
SparseVLM-v1.5Token Retention Rate=128, Base Model=LLaVA-1.5-7B2025.0668.57
VisPrunerToken Retention Rate=192, Base Model=LLaVA-1.5-7B2025.0668.52
VisionDropToken Retention Rate=128, Base Model=LLaVA-1.5-7B2025.0668.52
FastVToken Retention Rate=128, Base Model=LLaVA-1.5-7B2025.0668.37
ShortVBackbone=LLaVA-1.5-7B, TFLOPs=55%, N=202026.0668.27
FastVToken Retention Rate=192, Base Model=LLaVA-1.5-7B2025.0668.07
LLaVA-v1.5 (7B)Model=LLaVA-v1.5 (7B)2024.0566.8
MBQModel=Qwen2.5-VL-3B, Bits=W4A62026.0364.7
mPLUG-OWL2 (7B)Model=mPLUG-OWL2 (7B)2024.0564.5
InstructBLIP (7B)Model=InstructBLIP (7B)2024.0560.5
RTNModel=Qwen2.5-VL-3B, Bits=W4A82026.030