List your company
ResearchFormerly WizWand

Visual Question Answering on IconQA

87.2Top-1 AccFR-Merging
15.12833.83952.5571.261May 11, 2023Nov 2, 2023Apr 25, 2024Oct 17, 2024Apr 10, 2025Oct 2, 2025Mar 27, 2026
Updated 7mo ago

Evaluation results

MethodLinks
FR-MergingEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0387.2
EMR-MergingEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0384
RobustMergeEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0382.3
Qwen2.5-VL-3BTraining Stage=SCALe-GRPO2026.0375.58
Qwen2.5-VL-3BTraining Stage=GRPO2026.0374.26
Qwen2.5-VL-3BTraining Stage=SCALe-SFT2026.0371.91
Qwen2.5-VL-3BTraining Stage=SFT2026.0371.38
TARA-Variant BBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-based2026.0369.5
TARA-Variant ABackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-based2026.0369.2
LLaVA-v1.6-Mistral-7BTraining Stage=SCALe-GRPO2026.0369.17
Qwen2.5-VL-3BTraining Stage=Base2026.0369.04
AdaMergingEvaluation Protocol=Merged, Token Strategy=Full Token, Backbone=Vicuna 7B, LoRA rank=162026.0368.7
LLaVA-v1.6-Mistral-7BTraining Stage=SCALe-SFT2026.0368.25
LLaVA-v1.6-Mistral-7BTraining Stage=GRPO2026.0368.21
Gemma-3-4bTraining Stage=SCALe-GRPO2026.0367.8
FinetunedEvaluation Protocol=Finetuned, Backbone=Vicuna 7B, LoRA rank=162026.0367.8
LLaVA-v1.6-Mistral-7BTraining Stage=SFT2026.0367.47
TIESBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-free2026.0366.5
AdaMergingBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-based2026.0366.3
Gemma-3-4bTraining Stage=SCALe-SFT2026.0365.83
FinetunedBackbone=LLaVA, Adapter Rank=162026.0365.2
Gemma-3-4bTraining Stage=SFT2026.0364.8
AdaMergingEvaluation Protocol=Merged, Token Strategy=Single Token, Backbone=Vicuna 7B, LoRA rank=162026.0364.7
Gemma-3-4bTraining Stage=GRPO2026.0364.33
TIESEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0361.2
KnOTS-TIESBackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-free2026.0360.5
NSCEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0359.7
DARE-TIESEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0357.4
TABackbone=LLaVA, Adapter Rank=16, Optimization Strategy=gradient-free2026.0357.2
TAEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0356.8
SVDEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0356.2
LION-4Bparameters=4B2023.1154.91
LION-12Bparameters=12B2023.1154.89
InstructBLIP (T5XXL)backbone=T5XXL2023.1153.18
LoRA-LEGOEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0353.1
LinearEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0352.6
InstructBLIP (T5XL)backbone=T5XL2023.1152.47
KnOTS-DARE-TIESEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0352.4
InstructBLIPzero-shot=true, backbone=FlanT5-XXL2023.0551.2
InstructBLIP+ (T5XXL)backbone=T5XXL, uses_in-house_data=true2023.1151.2
InstructBLIPzero-shot=true, backbone=FlanT5-XL2023.0550
InstructBLIP+ (T5XL)backbone=T5XL, uses_in-house_data=true2023.1150
KnOTS-TIESEvaluation Protocol=Merged, Backbone=Vicuna 7B, LoRA rank=162026.0349.8
Pink2023.1147.8
MiniGPTV22023.1147.7
BLIP-2zero-shot=true, backbone=FlanT5-XL2023.0545.5
BLIP-2zero-shot=true, backbone=FlanT5-XXL2023.0545.4
BLIVA2023.1144.88
InstructBLIPzero-shot=true, backbone=Vicuna-13B2023.0544.8
InstructBLIPzero-shot=true, backbone=Vicuna-7B2023.0543.1
LLaVA2023.1143
BLIP-2zero-shot=true, backbone=Vicuna-13B2023.0540.6
BLIP-2zero-shot=true, backbone=Vicuna-7B2023.0539.7
MiniGPT42023.1137.6
LLaVA-v1.6-Mistral-7BTraining Stage=Base2026.0337.39
Gemma-3-4bTraining Stage=Base2026.0336.73
Zero-ShotEvaluation Protocol=Zero-Shot2026.0317.9