Datasets:

salma-remyx
/

Q-Spatial-Bench-sMAPE-Comparison

Model ID stringclasses 7 values	SpatialPrompt (P) stringclasses 2 values	Reasoning (R) stringclasses 2 values	Spatial FT (S) stringclasses 2 values	Model Family stringclasses 4 values	Param Count stringclasses 3 values	sMAPE int64 62 200
SpaceQwen2.5-VL-3B-Instruct	✅	❌	✅	Qwen2.5	3B	132
VLAA-Thinker-Qwen2.5VL-3B	✅	✅	❌	Qwen2.5	3B	130
gpt-4o	❌	✅	❌	OpenAI	Unknown	120
SpaceThinker-Qwen2.5VL-3B	❌	✅	✅	Qwen2.5	3B	95
SpaceQwen2.5-VL-3B-Instruct	❌	❌	✅	Qwen2.5	3B	94
gemini-2.5-pro-preview-03-25	❌	✅	✅	Gemini	Unknown	73
gpt-4o	✅	✅	❌	OpenAI	Unknown	71
SpaceThinker-Qwen2.5VL-3B	✅	✅	✅	Qwen2.5	3B	66
gemini-2.5-pro-preview-03-25	✅	✅	✅	Gemini	Unknown	62
SpaceLLaVA-7B	✅	❌	✅	LLaVA	7B	96
SpaceLLaVA-7B	❌	❌	✅	LLaVA	7B	200
SpaceThinker-7B	✅	✅	✅	LLaVA	7B	73
SpaceThinker-7B	❌	✅	✅	LLaVA	7B	85

Q-Spatial-Bench-sMAPE-Comparison

Recording sMAPE for models evaluated on Q-Spatial-Bench across model families, sizes, finetuning, prompting, post-training optimization strategies.