List your company

Visual Question Answering on GQA (Accuracy and Relative Performance)

63.9AccuracyVanilla
47.88452.04256.260.358Feb 3, 2026Feb 9, 2026Feb 15, 2026Feb 21, 2026Feb 27, 2026Mar 5, 2026Mar 11, 2026

Evaluation results

MethodLinks
VanillaToken Retention Ratio=100%, Backbone=LLaVA-NeXT-7B2026.0263.9100
SwiftVLMToken Retention Ratio=33.3%, Backbone=LLaVA-NeXT-7B2026.0263.698
SwiftVLMToken Retention Ratio=22.2%, Backbone=LLaVA-NeXT-7B2026.0263.597.1
LLaVA-1.5-VL-13BBackbone=LLaVA-1.5-VL-13B, Token=5762026.0363.3100
TPRLBackbone=LLaVA-1.5-VL-13B, Token=192*2026.0362.899.2
FEATHERToken Retention Ratio=33.3%, Backbone=LLaVA-NeXT-7B2026.0262.592.8
BaseBackbone=LLaVA-1.5-7B2026.0362—
FEATHERToken Retention Ratio=22.2%, Backbone=LLaVA-NeXT-7B2026.0261.987.5
VanillaVisual Token Budget=5762026.0261.4100
TPRLBackbone=LLaVA-1.5-VL-13B, Token=128*2026.0361.296.7
VisionZip+EEnergy Threshold (τ)=99.8%, Adaptive Budget (+E)=true, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=3122026.0360.998.2
VisionZipEnergy Threshold (τ)=99.8%, Adaptive Budget (+E)=false, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=3122026.0360.898.1
SwiftVLMVisual Token Budget=1922026.0260.798.9
PDropEnergy Threshold (τ)=99.8%, Adaptive Budget (+E)=false, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=3122026.0360.597.6
PDrop+EEnergy Threshold (τ)=99.8%, Adaptive Budget (+E)=true, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=3122026.0360.597.6
FastVEnergy Threshold (τ)=99.8%, Adaptive Budget (+E)=false, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=3122026.0360.497.4
FastV+EEnergy Threshold (τ)=99.8%, Adaptive Budget (+E)=true, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=3122026.0360.397.3
VisionZip+EEnergy Threshold (τ)=99.0%, Adaptive Budget (+E)=true, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=1772026.0359.596
SwiftVLMVisual Token Budget=1282026.0259.296.4
VisionZipEnergy Threshold (τ)=99.0%, Adaptive Budget (+E)=false, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=1772026.0359.195.3
FastVToken Retention Ratio=33.3%, Backbone=LLaVA-NeXT-7B2026.025975.1
PDrop+EEnergy Threshold (τ)=99.0%, Adaptive Budget (+E)=true, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=1772026.0358.394
PDropEnergy Threshold (τ)=99.0%, Adaptive Budget (+E)=false, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=1772026.0358.193.7
FastV+EEnergy Threshold (τ)=99.0%, Adaptive Budget (+E)=true, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=1772026.0357.592.7
FastVEnergy Threshold (τ)=99.0%, Adaptive Budget (+E)=false, Backbone=LLaVA-1.5-7B, Mean Effective Token Count=1772026.0357.492.6
FastVVisual Token Budget=1922026.0257.293.2
FastVToken Retention Ratio=22.2%, Backbone=LLaVA-NeXT-7B2026.0256.966.9
FastVBackbone=LLaVA-1.5-VL-13B, Token=1922026.0354.285.7
FastVBackbone=LLaVA-1.5-VL-13B, Token=1282026.0348.576.6