Fix Dataset Viewer config

README.md

@@ -21,6 +21,11 @@ language:
   - en
 size_categories:
   - n<1K
+configs:
+  - config_name: viewer_summary
+    data_files:
+      - split: train
+        path: viewer/dataset_viewer_summary.jsonl
 ---
 
 # Ropedia Xperience-10M Task Suite Artifacts

viewer/dataset_viewer_summary.jsonl

@@ -0,0 +1,9 @@
+{"row_type":"project_scope","title":"Public sample pipeline","value":"1 public Xperience-10M sample episode; 5,821 frames; 1,161 aligned 20-frame windows; 8,546-dimensional multimodal representation","source_file":"docs/data/project_status.json","notes":"This artifact repo stores derived outputs only and does not redistribute raw Xperience-10M MP4/HDF5/RRD data or full Qwen weights."}
+{"row_type":"modality_contract","title":"Modalities represented","value":"video, audio, depth, camera pose/SLAM, motion capture, inertial sensing, calibration metadata, and language annotations","source_file":"docs/data/xperience10m_dataset_card_alignment.json","notes":"Audio is featurized in the current public-sample task representation; raw synchronized data remains governed by the official upstream dataset terms."}
+{"row_type":"task_suite","title":"Embodied task suite","value":"12 human-readable task contracts with minimal baselines and compact neural MLP heads over the same chronological split","source_file":"results/episode_task_suite/summary_report.json","notes":"These are single-episode public-sample results, not held-out multi-episode model-quality claims."}
+{"row_type":"audio_study","title":"Audio contribution study","value":"Audio variants improve the primary metric on 6 of 12 task contracts in the current single-episode setting","source_file":"docs/data/project_status.json","notes":"See results/audio_ablation/ for the per-task audio/no-audio comparison and raw-logmel feature artifact."}
+{"row_type":"neural_baselines","title":"Compact neural heads","value":"12 PyTorch MLP task heads plus four extension probes are included as derived artifacts","source_file":"docs/data/project_status.json","notes":"The neural heads are local task baselines over derived features, separate from the pending Qwen3-Omni multi-episode pilot."}
+{"row_type":"official_upstream","title":"Official Xperience-10M dataset","value":"Official gated HF metadata observed: 12,103 annotation files, 12,102 complete visible episodes, 80,000+ listed files, and 31.9 TB live HF-hosted file-size display","source_file":"docs/data/xperience10m_dataset_card_alignment.json","notes":"The official dataset card also describes about 10M experience units, about 10,000 recording hours, and about 1 PB full-scale data; this artifact repo is not the raw dataset mirror."}
+{"row_type":"multi_episode_plan","title":"Selected multi-episode relay","value":"128 metadata-balanced episodes selected for future held-out training with a 96/16/16 train/val/test target split","source_file":"results/omni_finetune/DATA_ACCESS_STATUS.md","notes":"Staging, manifest construction, training, and held-out evaluation must complete before reporting full multi-episode metrics."}
+{"row_type":"foundation_model_plan","title":"Foundation-model branches","value":"Qwen3-Omni is the first trainable LoRA pilot; Cosmos 3 is planned as the world-model/action-generation branch; policy candidates include OpenVLA, openpi, and GR00T after action targets are explicit","source_file":"docs/data/project_status.json","notes":"The current package documents the plan and public-sample baselines; it does not claim a completed full-dataset fine-tune."}
+{"row_type":"viewer_config","title":"Dataset Viewer configuration","value":"This viewer_summary config points HF Dataset Viewer to one JSONL file so mixed artifact formats are not interpreted as incompatible splits","source_file":"viewer/dataset_viewer_summary.jsonl","notes":"The rest of the repo intentionally keeps CSV, JSON, JSONL, Markdown, HTML, NPZ, PT, PNG, and SVG artifacts for reproducibility and presentation."}