{"candidate": "ava_slowonly", "synthetic_input": true, "real_video_evaluation": false, "torch": "2.7.1+cu128", "cuda": "12.8", "gpu": "NVIDIA RTX PRO 4000 Blackwell", "input_shape": [1, 3, 4, 256, 256], "boxes_shape": [2, 5], "output_shape": [2, 81], "strict_load": true, "checkpoint_key_count": 320, "background_index": 0, "action_classes": 80, "compatibility": "Official backbone/head + torchvision RoIAlign; not full MMDetection wrapper", "roi_coordinate_contract": "PASS", "mmcv_numerical_equivalence": "NOT_TESTED", "status": "SMOKE_PASS", "wall_seconds": 1.533, "peak_cuda_allocated_bytes": 190102016, "peak_cuda_reserved_bytes": 278921216}
/opt/conda/lib/python3.11/site-packages/mmcv/cnn/bricks/transformer.py:33: UserWarning: Fail to import ``MultiScaleDeformableAttention`` from ``mmcv.ops.multi_scale_deform_attn``, You should install ``mmcv`` rather than ``mmcv-lite`` if you need this module. 
  warnings.warn('Fail to import ``MultiScaleDeformableAttention`` from '