List your company
ResearchFormerly WizWand

Visual Reasoning on TRACE (test)

0.346BLEU-1LTE-32B
0.087040.154270.22150.28873Oct 5, 2025
Updated 6mo ago

Evaluation results

MethodLinks
LTE-32BModel Scale=32B2025.100.3460.2110.327
LTE-7BModel Scale=7B2025.100.3140.2090.291
Largest 4 Bboxes-32BModel Scale=32B, Input Sampling Strategy=Largest 4 Bboxes2025.100.3030.1820.158
Largest 3 Bboxes-32BModel Scale=32B, Input Sampling Strategy=Largest 3 Bboxes2025.100.2990.1830.147
E-32B (one-turn)Model Scale=32B, Interaction Mode=one-turn, Output Configuration=explanation-only2025.100.2940.1530.315
E+G-32B (one-turn)Model Scale=32B, Interaction Mode=one-turn, Output Configuration=explanation and grounding-only2025.100.2820.1490.295
E-7B (one-turn)Model Scale=7B, Interaction Mode=one-turn, Output Configuration=explanation-only2025.100.280.1460.301
Dual Verdict Reward-32BModel Scale=32B, Training Configuration=Dual Verdict Reward2025.100.2760.1640.252
E+G-7B (one-turn)Model Scale=7B, Interaction Mode=one-turn, Output Configuration=explanation and grounding-only2025.100.2750.1360.269
No BLEU Reward-32BModel Scale=32B, Training Configuration=No BLEU Reward2025.100.1870.0950.153
SFT-32BModel Scale=32B, Training Stage=SFT2025.100.1590.0760.13
SFT-7BModel Scale=7B, Training Stage=SFT2025.100.1560.0420.129
Base-7BModel Scale=7B, Training Stage=Base (untrained)2025.100.110.0320.073
Random Cropping-32BModel Scale=32B, Input Sampling Strategy=Random Cropping2025.100.1050.0430.109
Base-32BModel Scale=32B, Training Stage=Base (untrained)2025.100.1020.0430.079
LEGION2025.100.1020.0580.054
FakeShield2025.100.0970.0560.067