diff options
Diffstat (limited to 'artifacts/vg_5k_tiers/manifest.json')
| -rw-r--r-- | artifacts/vg_5k_tiers/manifest.json | 18 |
1 files changed, 18 insertions, 0 deletions
diff --git a/artifacts/vg_5k_tiers/manifest.json b/artifacts/vg_5k_tiers/manifest.json new file mode 100644 index 0000000..d18db3c --- /dev/null +++ b/artifacts/vg_5k_tiers/manifest.json @@ -0,0 +1,18 @@ +{ + "dataset": "ranjaykrishna/visual_genome", + "seed": 20260728, + "selected_nodes": 5000, + "eligible_nodes": 107019, + "views_per_node": 16, + "minimum_available_regions": 32, + "extra_closure_tiers": true, + "nested_scaling_sizes": [ + 5000 + ], + "training_files": { + "vision": "vision_nodes.private.jsonl", + "text": "text_nodes.jsonl" + }, + "evaluation_only": "ground_truth.private.jsonl", + "leakage_note": "The source image ID and URL are required only by the vision feature preprocessor. Alignment training must consume extracted features with source metadata removed." +}
\ No newline at end of file |
