Dataset Monitor
Training data pipeline status and finetune controls.
Total records
389
Training
315
Validation
41
Test
33
Training threshold
ready_for_training: true ✓315 training recordsthreshold: 100
Mean composite score: 71%
AC-AMI signal gate threshold: 0.68. Records below this score are excluded from training splits.
By source
389
human
By subgenre
Private School
386
Stixx Sgija
3
Finetune controls
Triggers MusicGen fine-tuning on Modal (A10G GPU, ~45 min, ~$3). Only available when dataset is ready.
How to ingest tracks
# 1. Generate a track via the agent
POST /api/agent/run { title, subgenre, ... }
# 2. Ingest to dataset pipeline
POST /api/agent/ingest { track_id, generation_id, audio_url, source: "generated" }
# 3. Monitor workflow
GET /api/agent/workflow/:workflowId
# 4. Auto-trigger fires when ready
modal deploy apps/audio/modal_auto_trigger.py