List your company
ResearchFormerly WizWand

Science Question Answering on ScienceQA (Accuracy)

96.33AccuracyUDS
61.458870.511979.56588.6181Feb 17, 2025May 9, 2025Jul 30, 2025Oct 20, 2025Jan 10, 2026Apr 2, 2026Jun 23, 2026
Updated 4mo ago

Evaluation results

MethodLinks
UDSModel=Qwen-2.5-14B, Full Finetuning=true2025.1096.33
RegularModel=Qwen-2.5-14B, Full Finetuning=true2025.1095.86
UDSModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1095.56
GREATSModel=Qwen-2.5-14B, Full Finetuning=true2025.1095.54
RHO-LossModel=Qwen-2.5-14B, Full Finetuning=true2025.1095.25
GREATSModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1095.22
RHO-LossModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1095.14
MaxGradModel=Qwen-2.5-14B, Full Finetuning=true2025.1095.12
UDSModel=Qwen-2.5-7B, Full Finetuning=true2025.1095.06
MaxLossModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1094.92
RandomModel=Qwen-2.5-14B, Full Finetuning=true2025.1094.91
MaxLossModel=Qwen-2.5-14B, Full Finetuning=true2025.1094.85
GREATSModel=Qwen-2.5-7B, Full Finetuning=true2025.1094.75
RHO-LossModel=Qwen-2.5-7B, Full Finetuning=true2025.1094.64
RandomModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1094.6
RegularModel=Qwen-2.5-7B, Full Finetuning=true2025.1094.52
RegularModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1094.47
MaxGradModel=Qwen-2.5-7B-Instruct, Tuning Mode=Instruction Tuning2025.1094.46
MaxLossModel=Qwen-2.5-7B, Full Finetuning=true2025.1094.41
RandomModel=Qwen-2.5-7B, Full Finetuning=true2025.1094.38
MaxGradModel=Qwen-2.5-7B, Full Finetuning=true2025.1094.27
PRISM-QwenModel Size=7B, Backbone=Qwen, Training Strategy=Instruct2025.0282.3
PRISM-QwenModel Size=7B, Backbone=Qwen, Training Strategy=Base2025.0281.2
LLaVA-QwenModel Size=7B, Backbone=Qwen, Training Strategy=Instruct2025.0280.2
PRISMModel Size=8B, Backbone=Llama2025.0279.4
LLaVA-QwenModel Size=7B, Backbone=Qwen, Training Strategy=Base2025.0279.1
LLaVA-LlamaModel Size=8B, Backbone=Llama2025.0277.6
PRISMModel Size=3B, Backbone=Phi22025.0276.3
LLaVA-Phi2Model Size=3B, Backbone=Phi22025.0275.3
PRISMModel Size=13B, Backbone=Vicuna2025.0274.5
LLaVA-VicunaModel Size=13B, Backbone=Vicuna2025.0274.4
PRISMModel Size=7B, Backbone=Vicuna2025.0271.3
VanillaVisual Token Budget=2880, Token Retention Ratio=100%2026.0670.1
LLaVA-VicunaModel Size=7B, Backbone=Vicuna2025.0269.4
PriorTRVisual Token Budget=320, Token Retention Ratio=11.1%2026.0668.1
PDropVisual Token Budget=320, Token Retention Ratio=11.1%2026.0667.5
PruMergeVisual Token Budget=320, Token Retention Ratio=11.1%2026.0666.4
SparseVLMVisual Token Budget=320, Token Retention Ratio=11.1%2026.0666.1
FastVVisual Token Budget=320, Token Retention Ratio=11.1%2026.0662.8