Buckets:
| # Visual-OPSD main training config (paper default, on-policy). | |
| # Used by scripts/visual_opsd/run_visual_opsd.sh -> | |
| # scripts/visual_opsd/train_visual_opsd.py. | |
| # Each "visual_opsd" group is consumed by ``OPSDPairedIterableDataset`` | |
| # which emits raw envelopes (problem image + question + VT images + | |
| # ground-truth answer for logging). The on-policy trainer then samples | |
| # a completion from the student and packs the student/teacher batches | |
| # per step. | |
| visual_opsd: | |
| dataset_names: | |
| - Visual_Search | |
| - Spatial_Navigation | |
| - Jigsaw_Assembly | |
| - Chart_Refocus | |
| image_transform_args: | |
| image_stride: 14 | |
| max_image_size: 980 | |
| min_image_size: 378 | |
| max_pixels: 2_007_040 | |
| is_mandatory: true | |
| num_used_data: | |
| - 6990 | |
| - 6000 | |
| - 6000 | |
| - 6000 | |
| weight: 1 | |
Xet Storage Details
- Size:
- 767 Bytes
- Xet hash:
- 9adf91bfca637751a4cc01b836e0e6a120270ae6baa93e63e60ddb23157ff465
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.