| |
| """Example: Inspect timeline chunks in detail.""" |
| from datasets import load_dataset |
|
|
| REPO = "skyzhou06/StreamingOmniDatasets" |
| CONFIG = "AVJointReasoning" |
|
|
| print(f"Loading {REPO}/{CONFIG} eval split...") |
| try: |
| ds = load_dataset(REPO, CONFIG, split="eval", streaming=True) |
| sample = next(iter(ds)) |
| print(f"Row id: {sample.get('id')}") |
| print(f"Question: {sample.get('question')}") |
| print(f"Answer: {sample.get('answer')}") |
|
|
| timeline = sample.get("timeline", []) |
| print(f"\nTimeline ({len(timeline)} chunks):") |
| for i, chunk in enumerate(timeline[:5]): |
| print(f"\n Chunk {i}:") |
| print(f" t_start={chunk.get('t_start')}, t_end={chunk.get('t_end')}") |
| print(f" control={chunk.get('control')}") |
| asr = chunk.get('asr_delta') |
| if asr: |
| print(f" asr_delta={str(asr)[:80]}") |
| audio_events = chunk.get('audio_events') |
| if audio_events: |
| print(f" audio_events={audio_events}") |
| visual_events = chunk.get('visual_events') |
| if visual_events: |
| print(f" visual_events={visual_events}") |
| if len(timeline) > 5: |
| print(f"\n ... and {len(timeline)-5} more chunks") |
| except Exception as e: |
| print(f"Config-based loading failed: {e}") |
| print("Try parquet fallback as shown in load_streaming_omni_dataset.py") |
|
|
