Run Visual-Perception Evaluation
Launch evaluation jobs in a background process. Results are written to results/visual_perception/ and appear in the Overview page when complete.
Select Command
Prerequisites
- • evaluate / smoke-test: POPE dataset + Qwen2.5-VL-3B-Instruct weights
- • robustness / ablation: needs a completed evaluate run first
- • validate-data: checks datasets only, no model needed