List your company
ResearchFormerly WizWand

Visual Question Answering on SciQA IMG

87.1AccuracyASMv2
56.170464.200272.2380.2598Nov 7, 2023Apr 16, 2024Sep 24, 2024Mar 5, 2025Aug 13, 2025Jan 21, 2026Jul 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
ASMv2Model Scale=13B, Seen Training Images=true2024.0287.1
InfiMM-HDLLM=Vicuna-13B, In-house data=false2024.0383.6
GroundingGPTLLM Size=7B2024.0178.7
VILAModel Scale=13B2024.0273.7
LLaVA-Next 13BLLM=Vicuna-13B, Encoder=CLIP VIT-large/14, Resolution=672, Zero-shot=true2024.0473.57
FP16Model=LLaVA-Next 13B2026.0573.23
WSVDModel=LLaVA-Next 13B, Quantization=W8A8, rho_1=50%2026.0573.08
LLaVA-v1.5 13BLLM=Vicuna-13B, Encoder=CLIP VIT-large/14, Resolution=336, Zero-shot=true2024.0472.88
LASERModel=LLaVA-Next 13B, Quantization=W8A8, rho_1=50%2026.0572.11
LLaVA 1.5LLM=Vicuna-13B, In-house data=false2024.0371.6
LLaVA-1.5Model Scale=13B2024.0271.6
OmniFusionLLM=GigaChat-7B, Encoder=InternViT-6B-448px V1-2, Resolution=448, Zero-shot=true2024.0471.29
HyperLLaVA w/o ELLLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0370.7
Sphinx-2KLLM=LLAMA2-13B, In-house data=false2024.0370.6
LLaVA-v1.5 7BLLM=Vicuna-7B, Encoder=CLIP VIT-large/14, Resolution=336, Zero-shot=true2024.0470.43
QSVDModel=LLaVA-Next 13B, Quantization=W8A42026.0570.43
HyperLLaVA w/o EvLLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0370.4
HyperLLaVALLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0370.4
DuQuantModel=LLaVA-Next 13B, Quantization=W8A42026.0570.2
LLaVA-Next 7BLLM=Vicuna-7B, Encoder=CLIP VIT-large/14, Resolution=672, Zero-shot=true2024.0470.15
OmniFusionLLM=Vicuna-7B, Encoder=CLIP VIT-large/14 + DinoV2, Resolution=336, Zero-shot=true2024.0469.96
FP16Model=LLaVA-Next 7B2026.0569.6
LLaVA-1.5-7BRetain Tokens=576, Pruning Ratio=0%2026.0769.6
PDropRetain Tokens=128, Pruning Ratio=77.8%2026.0769.5
CDPrunerRetain Tokens=32, Pruning Ratio=94.4%2026.0769.5
MonkeyLLM=Qwen-7B, In-house data=true2024.0369.4
SparseVLMRetain Tokens=128, Pruning Ratio=77.8%2026.0769.3
EADPRetain Tokens=32, Pruning Ratio=94.4%2026.0769.3
PruMerge+Retain Tokens=64, Pruning Ratio=88.9%2026.0769.2
OmniFusionLLM=GigaChat-7B, Encoder=CLIP VIT-large/14 + grid-split, Resolution=672, Zero-shot=true2024.0469.16
DARTRetain Tokens=128, Pruning Ratio=77.8%2026.0769.1
TRIMRetain Tokens=64, Pruning Ratio=88.9%2026.0769.1
DARTRetain Tokens=32, Pruning Ratio=94.4%2026.0769.1
CDPrunerRetain Tokens=128, Pruning Ratio=77.8%2026.0769
EADPRetain Tokens=128, Pruning Ratio=77.8%2026.0769
SparseVLMRetain Tokens=64, Pruning Ratio=88.9%2026.0769
VisionZipRetain Tokens=64, Pruning Ratio=88.9%2026.0769
VisionZipRetain Tokens=128, Pruning Ratio=77.8%2026.0768.9
HiPruneRetain Tokens=64, Pruning Ratio=88.9%2026.0768.9
EADPRetain Tokens=64, Pruning Ratio=88.9%2026.0768.9
mPLUG-Owl2#Params=8.2B, Model Type=Generalist, Evaluation Protocol=Zero-shot2023.1168.7
mPLUG-Owl2LLM=LLaMA2-7B, In-house data=false2024.0368.7
FastVRetain Tokens=128, Pruning Ratio=77.8%2026.0768.7
FastVRetain Tokens=64, Pruning Ratio=88.9%2026.0768.7
DARTRetain Tokens=64, Pruning Ratio=88.9%2026.0768.7
VisionZipRetain Tokens=32, Pruning Ratio=94.4%2026.0768.7
PruMerge+Retain Tokens=128, Pruning Ratio=77.8%2026.0768.6
DivPruneRetain Tokens=32, Pruning Ratio=94.4%2026.0768.6
LASERModel=LLaVA-Next 7B, Quantization=W8A8, rho_1=50%2026.0568.37
TRIMRetain Tokens=128, Pruning Ratio=77.8%2026.0768.3
HiPruneRetain Tokens=128, Pruning Ratio=77.8%2026.0768.3
PDropRetain Tokens=64, Pruning Ratio=88.9%2026.0768.3
Qwen-VL-Chat#Params=9.6B, Model Type=Generalist, Evaluation Protocol=Zero-shot2023.1168.2
Qwen-VL-Chat#Params=9.6B, Evaluation Mode=Zero-shot2023.1268.2
Qwen-VL-ChatLLM=Qwen-7B, In-house data=true2024.0368.2
Qwen-VL-ChatLLM=Qwen-7B, Res.=448, PT=1.4B, IT=50M2024.0368.2
Qwen-VL-ChatLLM=Qwen-7B, Encoder=CLIP VIT-bigG/14, Resolution=448, Zero-shot=true2024.0468.2
Qwen-VL-Chat2024.0268.2
DivPruneRetain Tokens=128, Pruning Ratio=77.8%2026.0768.2
OmniFusionLLM=GigaChat-7B, Encoder=CLIP VIT-large/14, Resolution=336, Zero-shot=true2024.0468.17
FP16Model=LLaVA-v1.5 7B2026.0568.1
CDPrunerRetain Tokens=64, Pruning Ratio=88.9%2026.0768.1
PruMerge+Retain Tokens=32, Pruning Ratio=94.4%2026.0768.1
DivPruneRetain Tokens=64, Pruning Ratio=88.9%2026.0768
TRIMRetain Tokens=32, Pruning Ratio=94.4%2026.0768
OmniFusionLLM=Mistral-7B, Encoder=CLIP VIT-large/14 + DinoV2, Resolution=336, Zero-shot=true2024.0467.55
LLaVA1.5-HACL#Params=7.2B, Evaluation Mode=Zero-shot2023.1267.3
OmniFusionLLM=GigaChat-7B, Encoder=CLIP VIT-large/14 + DinoV2, Resolution=336, Zero-shot=true2024.0467.18
Qwen-VLLLM=Qwen-7B, In-house data=true2024.0367.1
Qwen-VLLLM=Qwen-7B, Res.=448, PT=1.4B, IT=50M2024.0367.1
Qwen-VL2024.0267.1
WSVDModel=LLaVA-Next 7B, Quantization=W8A8, rho_1=50%2026.0566.94
LLaVA1.5#Params=7.2B, Evaluation Mode=Zero-shot2023.1266.8
LLaVA-1.5LLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0366.8
LLaVA-1.5LLM Size=7B2024.0166.8
DuQuantModel=LLaVA-Next 7B, Quantization=W8A42026.0566.34
LASERModel=LLaVA-v1.5 7B, Quantization=W8A8, rho_1=50%2026.0566.14
QSVDModel=LLaVA-Next 7B, Quantization=W8A42026.0566.1
QSVDModel=LLaVA-v1.5 7B, Quantization=W8A42026.0565.61
QVLMModel=LLaVA-Next 13B, Quantization=W8A42026.0565.28
WSVDModel=LLaVA-v1.5 7B, Quantization=W8A8, rho_1=50%2026.0564.25
InstructBLIPLLM=Vicuna-13B, Res.=224, PT=129M, IT=1.2M2024.0363.1
InstructBLIPModel Scale=13B2024.0263.1
InstructBLIPLLM Size=13B2024.0163.1
LLaVA-HACL#Params=7.2B, Evaluation Mode=Zero-shot2023.1262.4
LLaVA-1.5#Params=7.2B, Model Type=Generalist, Evaluation Protocol=Zero-shot, OCR inputs=true2023.1161.5
LLaVA#Params=7.2B, Evaluation Mode=Zero-shot2023.1261.5
BLIP-2#Params=8.2B, Model Type=Generalist, Evaluation Protocol=Zero-shot2023.1161
BLIP-2#Params=8.2B, Evaluation Mode=Zero-shot2023.1261
BLIP-2LLM=Vicuna-13B, Res.=224, PT=129M, IT=-2024.0361
BLIP-22024.0261
BLIP-2LLM Size=13B2024.0161
QVLMModel=LLaVA-Next 7B, Quantization=W8A42026.0560.6
InstructBLIP#Params=8.2B, Model Type=Generalist, Evaluation Protocol=Zero-shot2023.1160.5
InstructBLIP#Params=8.2B, Evaluation Mode=Zero-shot2023.1260.5
InstructBLIPLLM=Vicuna-7B, Res.=224, PT=129M, IT=1.2M2024.0360.5
InstructBLIPLLM Size=7B2024.0160.5
MiniGPT-4-HACL#Params=7.2B, Evaluation Mode=Zero-shot2023.1260.3
MiniGPT-4#Params=7.2B, Evaluation Mode=Zero-shot2023.1258.4
DuQuantModel=LLaVA-v1.5 7B, Quantization=W8A42026.0557.36