List your company

Generative Hallucination on AMBER Generative

70.4Coverage (%)Qwen2.5-VL 3B + SFT
32.02441.98751.9561.913Mar 1, 2025May 15, 2025Jul 30, 2025Oct 14, 2025Dec 28, 2025Mar 14, 2026May 29, 2026

Evaluation results

MethodLinks
Qwen2.5-VL 3B + SFTModel=Qwen2.5-VL 3B, Training Strategy=+ SFT2025.1070.4—48.83.96.9——
Qwen2.5-VL 3BModel=Qwen2.5-VL 3B, Training Strategy=Base2025.1069.9—55.75.87.6——
Qwen2.5-VL 3B + GRPOModel=Qwen2.5-VL 3B, Training Strategy=+ GRPO2025.1068.5—44.64.26.7——
Qwen2.5-VL 3B + NoisyGRPOModel=Qwen2.5-VL 3B, Training Strategy=+ NoisyGRPO2025.1067.7—44.33.46.6——
GPT-4V (Base)Backbone=GPT-4V2025.0367.14.630.72.6———
GPT4VModel=GPT4V2025.1067.1—30.72.64.6——
Qwen2.5-VL 7B + SFTModel=Qwen2.5-VL 7B, Training Strategy=+ SFT2025.1064.9—32.42.15.4——
Qwen2.5-VL 7B + GRPOModel=Qwen2.5-VL 7B, Training Strategy=+ GRPO2025.1064.3—28.91.74.9——
Qwen2.5-VL 7BModel=Qwen2.5-VL 7B, Training Strategy=Base2025.1063.6—27.51.64.8——
Qwen2.5-VL 7B + NoisyGRPOModel=Qwen2.5-VL 7B, Training Strategy=+ NoisyGRPO2025.1063.2—23.91.24.2——
CogVLM-ChatModel=CogVLM-Chat2025.1057.2—23.61.35.6——
VCDBase Model=LLaVA-Next-7B2026.0456.17.136.33———
LLaVA-Next-7BBase Model=LLaVA-Next-7B2026.0455.87.137.63.1———
M3IDBase Model=LLaVA-Next-7B2026.0455.66.935.83.3———
AVISCBase Model=LLaVA-Next-7B2026.0455.27.5382.5———
AttnRealBase Model=InstructBLIP2026.0554.35.730.52.9—27.2—
AttnRealBase Model=Qwen-VL2026.0554.25.325.92.1—0.7—
VCDBase Model=InstructBLIP2026.0554.28.339.13.7—2.1—
VanillaBase Model=InstructBLIP2026.0553.88.135.33.7—2.3—
VanillaBase Model=Qwen-VL2026.0553.36.329.62.6—0.2—
HALVABackbone=LLaVA-v1.5-7B2025.03536.632.23.4———
VCD + CNSenhancement=Contrastive Neuron Steering2026.0152.86.934.23.7———
Vanilla + CNSenhancement=Contrastive Neuron Steering2026.0152.47.134.73.9———
VCDBase Model=Shikra2026.0552.310.7504.9—0.2—
PSRDBase Model=LLaVA-1.5-13B2026.04524.724.12.2———
CASBase Model=InstructBLIP2026.0551.96.128.82.8—2.6—
PADE2026.0251.87.131.43.4———
AttnRealBase Model=Shikra2026.0551.78.743.34.3—0.5—
DoLA2026.0251.67.6364———
VanillaBase Model=Shikra2026.0551.410.848.55.5—0.3—
Vanilla2026.01517.836.44.2———
Vanilla2026.02517.836.44.2———
LLaVA-1.5 7BModel=LLaVA-1.5 7B2025.1051—36.44.27.8——
VCD2026.0150.87.436.14———
VCD2026.0250.87.536.24.1———
VCDBase Model=Qwen-VL2026.0550.88.837.53.4—0.2—
PAIBase Model=LLaVA-1.52026.0550.54.423.21.6—29—
CASBase Model=Shikra2026.0550.57.634.22.9—0.1—
VCDBase Model=LLaVA-1.52026.0550.48.736.23.9—0.5—
AttnRealBase Model=LLaVA-1.52026.0550.25.3262.4—4.9—
CASBase Model=Qwen-VL2026.0550.25.323.51.9—0.5—
PAI2026.0249.97.433.23.7———
HA-DPOBackbone=LLaVA-v1.5-7B2025.0349.86.730.93.3———
OPERA2026.0249.67.3323.5———
AVISCBase Model=LLaVA-1.5-13B2026.0449.58.639.63.6———
Qwen-VL-ChatModel=Qwen-VL-Chat2025.1049.4—23.61.95.5——
LLaVA-1.5-13BBase Model=LLaVA-1.5-13B2026.0449.46.728.83.1———
VanillaBase Model=LLaVA-1.52026.0549.47.731.13.6—0.5—
OctopusBackbone=LLaVA-v1.5-7B2025.0349.24.823.41.2———
EOSBackbone=LLaVA-v1.5-7B2025.0349.15.122.72———
M3IDBackbone=LLaVA-v1.5-7B, reproduced=true2025.0348.96261.5———
YARDBackbone=LLaVA-1.5-7B2026.0548.94.621.61.2———
M3IDBackbone=LLaVA-1.5-7B2026.0548.75.425.51.4———
OctopusBackbone=Instruct-BLIP2025.0348.56.122.21.3———
OctopusBase Model=InstructBLIP-7B2026.0448.56.122.21.3———
VCDBase Model=LLaVA-1.5-13B2026.0448.36.733.63.5———
FuzzyCDBackbone=LLaVA-1.5-7B2026.0548.16.227.31.8———
PSRDBase Model=InstructBLIP-7B2026.0447.84.420.91.8———
VCDBackbone=Instruct-BLIP, reproduced=true2025.0347.77.629.92.2———
VCDBase Model=InstructBLIP-7B2026.0447.77.629.92.2———
M3IDBase Model=LLaVA-1.5-13B2026.0447.7630.43.3———
M3IDBackbone=Instruct-BLIP, reproduced=true2025.0347.26.927.52.2———
M3IDBase Model=InstructBLIP-7B2026.0447.26.927.52.2———
AVISCBackbone=Instruct-BLIP, reproduced=true2025.0346.76.7282.6———
AVISCBase Model=InstructBLIP-7B2026.0446.76.7282.6———
CASBase Model=LLaVA-1.52026.0546.74.417.21.6—0.1—
VCDBackbone=LLaVA-1.5-7B2026.0546.76.125.82.1———
AVISCBackbone=LLaVA-v1.5-7B, reproduced=true2025.0346.66.325.62———
VCDBackbone=LLaVA-v1.5-7B, reproduced=true2025.0346.56.727.82———
Base (Comparable)Backbone=Instruct-BLIP2025.0346.48.431.12.6———
InstructBLIP-7BBase Model=InstructBLIP-7B2026.0446.48.431.12.6———
ICDBackbone=LLaVA-1.5-7B2026.0546.36.325.82.2———
EVASBackbone=LLaVA-1.5-7B2026.0546.14.718.81.6———
OPERABackbone=LLaVA-1.5-7B2026.0545.64.618.91.6———
PSRDBase Model=LLaVA-Next-7B2026.0445.34.721.11.2———
AVISCBackbone=LLaVA-1.5-7B2026.0545.27.127.62.1———
LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.05458.3322.2———
Base (Comparable)Backbone=LLaVA-v1.5-7B2025.0344.58312.2———
SSLBase Model=LLaVA-1.52026.0544.28.331.63.4—0.4—
TAMEBackbone=LLaVA-1.5-7B2026.0540.61235.63.4———
PAIBase Model=Shikra2026.0533.58.221.41.9—1.4—
SFT + DA-DPOBase Model=Qwen2.5-VL-7B2026.06——————44.8
SFT + DA-DPOBase Model=Qwen2.5-VL-72B2026.06——————49.8
SFT + DPOBase Model=Qwen2.5-VL-7B2026.06——————42.5
SFT + DPOBase Model=Qwen2.5-VL-72B2026.06——————46.8
SFT + HA-DPOBase Model=Qwen2.5-VL-72B2026.06——————48.5
SFT + SimPOBase Model=Qwen2.5-VL-7B2026.06——————42.8
SFT + SimPOBase Model=Qwen2.5-VL-72B2026.06——————47
SFT + VIGILBase Model=Qwen2.5-VL-7B2026.06——————46.5
SFT + VIGILBase Model=Qwen2.5-VL-72B2026.06——————52.2
SFT onlyBase Model=Qwen2.5-VL-7B2026.06——————41.2
SFT onlyBase Model=Qwen2.5-VL-72B2026.06——————45.3