List your company
ResearchFormerly WizWand

Multi-modal Understanding on MMBench EN (Accuracy)

93.53AccuracyMaLoRA
21.322840.068958.81577.5611Dec 6, 2024Mar 7, 2025Jun 7, 2025Sep 6, 2025Dec 7, 2025Mar 8, 2026Jun 8, 2026
Updated 4mo ago

Evaluation results

MethodLinks
MaLoRAModel=Qwen3-VL-8B, Training setting=Ours, Training examples=3.5k2025.1093.53
MaLoRAModel=Qwen2.5-VL-7B, Training setting=Ours, Training examples=3.5k2025.1092.84
LoRAModel=Qwen3-VL-8B, Training setting=LoRA, Training examples=3.5k2025.1092.03
BaseModel=Qwen3-VL-8B, Training setting=Base, Training examples=3.5k2025.1091.34
BaseModel=Qwen2.5-VL-7B, Training setting=Base, Training examples=3.5k2025.1090.3
LoRAModel=Qwen2.5-VL-7B, Training setting=LoRA, Training examples=3.5k2025.1090.18
QLoRAModel=Qwen3-VL-8B, Training setting=QLoRA, Training examples=3.5k2025.1089.03
QLoRAModel=Qwen2.5-VL-7B, Training setting=QLoRA, Training examples=3.5k2025.1088.91
Ours:base Qwen2.5Params (B)=14+72, Time (s)=3.22025.0888.54
InternVL2.5-78BModel Name=InternVL2.5-78B2024.1288.3
InternVL2-40BModel Name=InternVL2-40B2024.1286.8
Qwen2.5-VLParams (B)=722025.0886.61
InternVL2.5-38BModel Name=InternVL2.5-38B2024.1286.5
Qwen2-VL-72BModel Name=Qwen2-VL-72B2024.1286.5
InternVL2-Llama3-76BModel Name=InternVL2-Llama3-76B2024.1286.5
LLaVA-OVParams (B)=722025.0885.9
LLaVA-OneVision-72BModel Name=LLaVA-OneVision-72B2024.1285.8
Qwen2.5-VLParams (B)=322025.0885.55
InternVL3.5-38BRatio=100%, Backbone=InternVL3.5-38B2026.0585.47
InternVL2.5-26BModel Name=InternVL2.5-26B2024.1285.4
F3ARatio=60%, Backbone=InternVL3.5-38B2026.0585.21
VisionZipRatio=60%, Backbone=InternVL3.5-38B2026.0585.01
SenseNova-SIBase Architecture=InternVL3-8B, Model Category=Ours2025.1184.9
FastVRatio=60%, Backbone=InternVL3.5-38B2026.0584.79
InternVL2.5-8B2025.1284.6
InternVL2.5-8BModel Name=InternVL2.5-8B2024.1284.6
Qwen3-VL-8BModel Category=Open-source General Models2025.1184.6
DivPruneRatio=60%, Backbone=InternVL3.5-38B2026.0584.55
Qwen2.5-VLParams (B)=72025.0884.45
CDPrunerRatio=60%, Backbone=InternVL3.5-38B2026.0584.36
VisionZipRatio=40%, Backbone=InternVL3.5-38B2026.0584.01
FastVRatio=40%, Backbone=InternVL3.5-38B2026.0583.93
F3ARatio=40%, Backbone=InternVL3.5-38B2026.0583.93
DivPruneRatio=40%, Backbone=InternVL3.5-38B2026.0583.76
Qwen2.5-VL-7B2025.1283.5
SenseNova-SIBase Architecture=Qwen3-VL-8B, Model Category=Ours2025.1183.5
CDPrunerRatio=40%, Backbone=InternVL3.5-38B2026.0583.42
InternVL2-26BModel Name=InternVL2-26B2024.1283.4
GPT-4o-20240513Model Name=GPT-4o-202405132024.1283.4
SenseNova-SIBase Architecture=Bagel-7B-MoT, Model Category=Ours2025.1183.4
InternVL-2Params (B)=262025.0883.4
VST-7B-SFTModel Category=Open-source SI Models2025.1183.3
GPT-4oTime (s)=1.22025.0883.1
Qwen2-VL-7B2025.1283
Qwen2-VL-7BModel Name=Qwen2-VL-7B2024.1283
Bagel-7B-MoTModel Category=Open-source General Models2025.1182.8
MaLoRAModel=LLaVA-1.5-7B, Training setting=Ours, Training examples=3.5k2025.1082.8
Claude-3.5-SonnetModel Name=Claude-3.5-Sonnet2024.1282.6
InternVL-Chat-V1.5Model Name=InternVL-Chat-V1.52024.1282.2
F3ARatio=20%, Backbone=InternVL3.5-38B2026.0581.92
CoT4DET-7B2025.1281.9
Qwen2.5-OmniParams (B)=7, Time (s)=6.02025.0881.8
CDPrunerRatio=20%, Backbone=InternVL3.5-38B2026.0581.8
VisionZipRatio=20%, Backbone=InternVL3.5-38B2026.0581.73
InternVL2-8B2025.1281.7
InternVL2-8BModel Name=InternVL2-8B2024.1281.7
InternVL3-8BModel Category=Open-source General Models2025.1181.7
InternVL-2Params (B)=82025.0881.7
DivPruneRatio=20%, Backbone=InternVL3.5-38B2026.0581.62
MiniCPM-V2.6Model Name=MiniCPM-V2.62024.1281.5
FastVRatio=20%, Backbone=InternVL3.5-38B2026.0581.37
InternVL2.5-4BModel Name=InternVL2.5-4B2024.1281.1
GPT-4VModel Name=GPT-4V2024.1281
QLoRAModel=LLaVA-1.5-7B, Training setting=QLoRA, Training examples=3.5k2025.1080.95
LLaVA-OV-7B2025.1280.8
LLaVA-OVParams (B)=72025.0880.8
Cambrian-34BModel Name=Cambrian-34B2024.1280.4
Cambrian-S-7BModel Category=Open-source SI Models2025.1180.4
LoRAModel=LLaVA-1.5-7B, Training setting=LoRA, Training examples=3.5k2025.1080.25
InternVL3-2BModel Category=Open-source General Models2025.1179.7
SenseNova-SIBase Architecture=InternVL3-2B, Model Category=Ours2025.1178.9
InternVL2-4BModel Name=InternVL2-4B2024.1278.6
Qwen-VL-Max2025.0877.6
Phi-3.5-Vision-4BModel Name=Phi-3.5-Vision-4B2024.1276
GPT-4V2025.0875
Qwen2-VL-2BModel Name=Qwen2-VL-2B2024.1274.9
InternVL2.5-2BModel Name=InternVL2.5-2B2024.1274.7
DPVR-PCTrainable=202M, evaluation=3-seed average2026.0674.2
DPVR-KVTrainable=202M2026.0674.1
Gemini-1.5-ProModel Name=Gemini-1.5-Pro2024.1273.9
DPVR-LFTrainable=202M, evaluation=3-seed average2026.0673.8
Vanilla LLaVA-1.5-7BTrainable=02026.0673.7
LoRATrainable=80M, r=642026.0673.4
FastVTrainable=0, mode=inference-only2026.0673.3
InternVL2-2BModel Name=InternVL2-2B2024.1273.2
LoRATrainable=20M, r=162026.0672.7
BaseModel=LLaVA-1.5-7B, Training setting=Base, Training examples=3.5k2025.1072.52
VITA1.5Params (B)=7, Time (s)=3.72025.0871.8
InternVL2.5-1BModel Name=InternVL2.5-1B2024.1270.7
VanillaBackbone=LLaVA-v1.5-13B2025.1168.3
FullFTTrainable=7B2026.0666
InternVL2-1BModel Name=InternVL2-1B2024.1265.4
VanillaBackbone=LLaVA-v1.5-7B2025.1164.2
Claude-3-OpusModel Name=Claude-3-Opus2024.1263.3
LLaVA-OneVision-0.5BModel Name=LLaVA-OneVision-0.5B2024.1261.6
KL+PDBackbone=LLaVA-v1.5-7B2025.1132.7
SINEPROJECT(PO+PD)Backbone=LLaVA-v1.5-7B2025.1126.4
PO+PDBackbone=LLaVA-v1.5-7B2025.1126
KL+PDBackbone=LLaVA-v1.5-13B2025.1124.7
SINEPROJECT(PO+PD)Backbone=LLaVA-v1.5-13B2025.1124.1