List your company
ResearchFormerly WizWand

Document-Oriented Visual Question Answering on DocVQA

94.9AccuracyVanilla
-2.13223.05948.2573.441May 28, 2024Sep 7, 2024Dec 19, 2024Apr 1, 2025Jul 12, 2025Oct 23, 2025Feb 3, 2026
Updated 6mo ago

Evaluation results

MethodLinks
VanillaAvg. Tokens=100%2026.0294.9
IVC-PruneAvg. Tokens=50%2026.0294.4
FastVAvg. Tokens=52%2026.0293.9
PDropAvg. Tokens=52%2026.0293.8
LLaVA-OneVisionSize=72B2024.0993.1
MaLoRAModel=Qwen3-VL-8B, Training examples=39.5k2025.1092.88
Oryx-1.5Size=32B2024.0992.7
BaseModel=Qwen3-VL-8B, Training examples=39.5k2025.1092.67
LoRAModel=Qwen3-VL-8B, Training examples=39.5k2025.1092.54
QLoRAModel=Qwen3-VL-8B, Training examples=39.5k2025.1092.05
MaLoRAModel=Qwen2.5-VL-7B, Training examples=39.5k2025.1091.83
LoRAModel=Qwen2.5-VL-7B, Training examples=39.5k2025.1091.64
OryxSize=34B2024.0991.4
QLoRAModel=Qwen2.5-VL-7B, Training examples=39.5k2025.1090.3
Oryx-1.5Size=7B2024.0990.1
OryxSize=7B2024.0989
Florence-VL 8B# Vis tok.=5762024.1284.9
Magma-8B (Full)Number of Parameters=8B, Finetuning=true, Pre-training configuration=Full with SoM/ToM2025.0284.8
FastVLM-1.7BVision Encoder=FastViTHD, LLM=Qw.2, #Visual Tokens=256, Vis. Enc. Size (M)=125, Evaluation Library=lmms-eval2024.1284.2
Magma-8B (Actw/o)Number of Parameters=8B, Finetuning=true, Pre-training configuration=Action without SoM/ToM2025.0284.1
LLaVA-NeXTSize=34B2024.0984
Magma-8B (Fullw/o)Number of Parameters=8B, Finetuning=true, Pre-training configuration=Full without SoM/ToM2025.0283.8
Florence-VL 3B# Vis tok.=5762024.1282.1
FlorenceVL-3BVision Encoder=DaViT, LLM=Phi-3, #Visual Tokens=576, Vis. Enc. Size (M)=770, Evaluation Library=lmms-eval2024.1282.1
MM1.5-1BVision Encoder=ViT-H, #Visual Tokens=1440, Vis. Enc. Size (M)=632, Evaluation Library=lmms-eval2024.1281
Magma-8B (SFT)Number of Parameters=8B, Finetuning=true, Pre-training configuration=SFT2025.0280.4
LLaVA-NeXT+CALLLM=Vicuna-13B, Resolution setting=High resolution2024.0580.1
SmolVLM2-2.2BVision Encoder=ViT-SO400M, LLM=SmolLM2, #Visual Tokens=2106*, Vis. Enc. Size (M)=430, Evaluation Library=lmms-eval2024.1280
FastVLM-0.6BVision Encoder=FastViTHD, LLM=Qw.2, #Visual Tokens=256, Vis. Enc. Size (M)=125, Evaluation Library=lmms-eval2024.1279.1
LLaVA-NeXTLLM=Vicuna-13B, Resolution setting=High resolution2024.0578.4
BaseModel=Qwen2.5-VL-7B, Training examples=39.5k2025.1078.24
LLaVA-NeXTSize=8B2024.0978.2
MGM-HD+CALLLM=Vicuna-13B, Resolution setting=High resolution2024.0578
Cambrian 8B# Vis tok.=5762024.1277.8
Cambrian-1Size=8B2024.0977.8
MGM-HDLLM=Vicuna-13B, Resolution setting=High resolution2024.0577.7
MolmoE A1B-7BVision Encoder=ViT-SO400M, LLM=OLMOE., #Visual Tokens=1728, Vis. Enc. Size (M)=430, Evaluation Library=lmms-eval2024.1277.7
LLaVA-NeXT+CALLLM=Vicuna-7B, Resolution setting=High resolution2024.0577.3
Phi 3.5 Vision2024.1275.9
Cambrian-1Size=34B2024.0975.5
LLaVA-NeXTLLM=Vicuna-7B, Resolution setting=High resolution2024.0575.1
Mini-Gemini-HD 8B# Vis tok.=28802024.1274.6
LLaVA-Next-7BNumber of Parameters=7B, Finetuning=true2025.0274.4
MGM-HD+CALLLM=Vicuna-7B, Resolution setting=High resolution2024.0573.4
MGM-HDLLM=Vicuna-7B, Resolution setting=High resolution2024.0572
SmolVLM2-0.5BVision Encoder=ViT-B, LLM=SmolLM2, #Visual Tokens=1088*, Vis. Enc. Size (M)=93, Evaluation Library=VLMEvalKit2024.1270.5
LLaVA next 8B# Vis tok.=28802024.1269.3
MGMLLM=Vicuna-13B, Resolution setting=Low resolution2024.0561.7
MGM+CALLLM=Vicuna-13B, Resolution setting=Low resolution2024.0561.6
MGM+CALLLM=Vicuna-7B, Resolution setting=Low resolution2024.0558
MGMLLM=Vicuna-7B, Resolution setting=Low resolution2024.0557.7
MonkeyModel Generation Capability=Text only2024.0750.1
TextHarmony-ChatModel Generation Capability=Text and image, Slide-LoRA=true2024.0749.8
DocPediaModel Generation Capability=Text only2024.0747.1
TextHarmonyModel Generation Capability=Text and image, Slide-LoRA=true2024.0747.1
MGM+CALLLM=Gemma-2B, Resolution setting=Low resolution2024.0544.8
TextHarmony*Model Generation Capability=Text and image, Slide-LoRA=false2024.0744
MGMLLM=Gemma-2B, Resolution setting=Low resolution2024.0539.8
InternLM-XComposer2Model Generation Capability=Text only2024.0739.7
LLaVA-1.5+CALLLM=Vicuna-13B, Resolution setting=Low resolution2024.0532.6
LLaVA-1.5LLM=Vicuna-13B, Resolution setting=Low resolution2024.0531.1
LLaVA-1.5+CALLLM=Vicuna-7B, Resolution setting=Low resolution2024.0530.6
AIMv2architecture=ViT-3B/14, #patches=576, resolution=336px2024.1130.4
InternVLModel Generation Capability=Text only2024.0728.7
LLaVA-1.5LLM=Vicuna-7B, Resolution setting=Low resolution2024.0528.5
AIMv2architecture=ViT-1B/14, #patches=576, resolution=336px2024.1128.5
LLaVA-1.5-7BNumber of Parameters=7B, Finetuning=true2025.0228.1
AIMv2architecture=ViT-H/14, #patches=576, resolution=336px2024.1127.8
AIMv2architecture=ViT-L/14, #patches=576, resolution=336px2024.1126.6
OpenAI CLIParchitecture=ViT-L/14, #patches=5762024.1125.6
MaLoRAModel=LLaVA-1.5-7B, Training examples=39.5k2025.1025.37
LoRAModel=LLaVA-1.5-7B, Training examples=39.5k2025.1024.45
QLoRAModel=LLaVA-1.5-7B, Training examples=39.5k2025.1023.7
BaseModel=LLaVA-1.5-7B, Training examples=39.5k2025.1021.82
SigLIParchitecture=ViT-So400M/14, #patches=7522024.1119.2
mPLUG-Owl2Model Generation Capability=Text only2024.0717.9
SigLIParchitecture=ViT-L/14, #patches=5762024.1116.9
LLaVARModel Generation Capability=Text only2024.0712.3
LLaVA1.5-7BModel Generation Capability=Text only2024.078.5
DINOv2architecture=ViT-g/14, #patches=30342024.118.2
MM-InterleavedModel Generation Capability=Text and image2024.078.1
UniDocModel Generation Capability=Text only2024.077.7
SEED-LLaMA-14BModel Generation Capability=Text and image2024.076.5
MiniGPT5Model Generation Capability=Text and image2024.071.6