schema_version: "1.0" module: name: cornucopia version: "0.1.0" display_name: "CORNUCOPIA — Large-Scale Synthetic Data for VLA Pre-Training" description: "Scalable synthetic data generation pipeline for vision-language-action model pre-training" category: training.distillation license: Apache-2.0 paper: "Large-Scale Synthetic Data for VLA Pre-Training" capabilities: provides: - type: data_generation subtype: synthetic_vla - type: augmentation subtype: large_scale interface: inputs: - name: config ros2_type: std_msgs/msg/String outputs: - name: dataset_manifest ros2_type: std_msgs/msg/String hardware: platforms: - name: linux_x86 backends: [onnxruntime_cuda, onnxruntime_cpu] performance: profiles: - platform: linux_x86 model: cornucopia-base backend: onnxruntime_cuda fps: 5 latency_p50_ms: 200 safety: failure_mode: returns_empty timeout_ms: 60000 health_topic: /anima/cornucopia/health composability: pairs_well_with: - manipulation.vla - simulation.generation conflicts_with: [] container: image: ghcr.io/robotflow-labs/anima-cornucopia:0.1.0 port: 8085