LEGO Assembly Generation on Pun (test) benchmark leaderboard · SOTA2 Research

LEGO Assembly Generation on Pun (test)

0.77Qwen-VL ScorePVPO

Evaluation Results

Method	Links
PVPO2026.05		0.77	0.28	0.8	0.93	0.35	146
High-Value VLM + Diversity2026.05		0.74	0.27	0.76	0.89	0.32	181
High-Value VLM + Diversity2026.05		0.72	0.26	0.7	0.86	0.31	184
High-Value VLM2026.05		0.7	0.27	0.72	0.86	0.3	162
High-Value VLM2026.05		0.7	0.27	0.72	0.8	0.26	205
Full dataset2026.05		0.67	0.27	0.74	0.96	0.35	177
PVPO2026.05		0.67	0.27	0.74	0.97	0.35	179
Lowest perplexity2026.05		0.64	0.27	0.74	0.97	0.3	140
Full dataset2026.05		0.59	0.26	0.67	0.93	0.32	196
Diversity-only2026.05		0.58	0.26	0.66	0.95	0.32	163
Random subset2026.05		0.58	0.25	0.67	0.95	0.32	194
Random subset2026.05		0.56	0.25	0.64	0.91	0.28	176
Diversity-only2026.05		0.55	0.25	0.66	0.94	0.31	199
Low-value VLM2026.05		0.51	0.22	0.64	0.85	0.25	144
Shortest responses2026.05		0.5	0.24	0.65	0.91	0.22	33
Shortest responses2026.05		0.49	0.24	0.62	0.89	0.22	38
Longest responses2026.05		0.48	0.25	0.62	0.91	0.26	351
Lowest perplexity2026.05		0.45	0.25	0.56	0.88	0.22	136
Longest responses2026.05		0.44	0.23	0.57	0.86	0.26	346
Low-value VLM2026.05		0.28	0.22	0.57	0.87	0.29	334