Share your thoughts, 1 month free Claude Pro on usSee more

Visual Question Answering on BLINK

66.1Accuracy

CLAIMDIFF-RL actor-only

Updated 13d ago

Evaluation Results

Method	Links
CLAIMDIFF-RL actor-only 2026.05		66.1
HOLISTIC-RL 2026.05		65.8
CLAIMDIFF-RL relative 2026.05		65.28
Qwen3VL-32B-Instruct 2026.05		64.9
HOLISTIC-RL 2026.05		64.3
CLAIMDIFF-RL actor-only 2026.05		63.76
SFT 2026.05		61.6
GDRM 2026.04		48.39
GDRM 2026.04		48.29
Rotation 2025.10		48.18
GDRM 2026.04		47.71
GDRM 2026.04		47.66
Beam Search 2026.04		47.4
η-Sampling 2026.04		47.13
ε-Sampling 2026.04		46.61
GDRM 2026.04		46.45
Position 2025.10		46.39
Temp’ Only 2026.04		46.08
GDRM 2026.04		45.98
Greedy 2026.04		45.71
GDRM 2026.04		45.4
Min-p 2026.04		45.29
Jigsaw 2025.10		45.18
Top-p (nucleus) 2026.04		45.13
Contrastive 2025.10		45.13
Top-k 2026.04		44.77
Qwen2.5-VL-3B 2025.10		42.13