Training Throughput Measurement on LIBERO
1,620.39Training ThroughputRollout Async
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Rollout AsyncVLA Backbone=GR00T N1.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 1,620.39 | — | |
| StreamerVLA Backbone=GR00T N1.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 1,592.4 | 43.96 | |
| ColocatedVLA Backbone=GR00T N1.5, Number of GPUs=32 GPUs2026.02 | 1,125.62 | — | |
| StreamerVLA Backbone=pi0.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 1,120.91 | 59.25 | |
| Rollout AsyncVLA Backbone=pi0.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 1,041.36 | — | |
| Train AsyncVLA Backbone=GR00T N1.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 951.33 | — | |
| Rollout AsyncVLA Backbone=GR00T N1.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 816.48 | — | |
| StreamerVLA Backbone=GR00T N1.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 816.48 | 19.99 | |
| Train AsyncVLA Backbone=pi0.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 737.46 | — | |
| DisaggregatedVLA Backbone=GR00T N1.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 729.98 | — | |
| StreamerVLA Backbone=pi0.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 713.38 | 30.29 | |
| ColocatedVLA Backbone=pi0.5, Number of GPUs=32 GPUs2026.02 | 703.85 | — | |
| Rollout AsyncVLA Backbone=pi0.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 686.8 | — | |
| ColocatedVLA Backbone=GR00T N1.5, Number of GPUs=16 GPUs2026.02 | 680.46 | — | |
| ColocatedVLA Backbone=pi0.5, Number of GPUs=16 GPUs2026.02 | 547.55 | — | |
| Train AsyncVLA Backbone=GR00T N1.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 477.2 | — | |
| DisaggregatedVLA Backbone=pi0.5, Number of GPUs=32 GPUs, GPU Allocation Ratio=1:12026.02 | 457.23 | — | |
| Train AsyncVLA Backbone=pi0.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 441.49 | — | |
| StreamerVLA Backbone=GR00T N1.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 439.64 | 18.25 | |
| Rollout AsyncVLA Backbone=GR00T N1.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 434.2 | — | |
| DisaggregatedVLA Backbone=GR00T N1.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 409.6 | — | |
| StreamerVLA Backbone=pi0.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 383.4 | 32.5 | |
| ColocatedVLA Backbone=GR00T N1.5, Number of GPUs=8 GPUs2026.02 | 371.8 | — | |
| Rollout AsyncVLA Backbone=pi0.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 369.56 | — | |
| DisaggregatedVLA Backbone=pi0.5, Number of GPUs=16 GPUs, GPU Allocation Ratio=1:12026.02 | 307.84 | — | |
| ColocatedVLA Backbone=pi0.5, Number of GPUs=8 GPUs2026.02 | 289.23 | — | |
| Train AsyncVLA Backbone=GR00T N1.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 243.57 | — | |
| Train AsyncVLA Backbone=pi0.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 229.68 | — | |
| DisaggregatedVLA Backbone=GR00T N1.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 220.81 | — | |
| DisaggregatedVLA Backbone=pi0.5, Number of GPUs=8 GPUs, GPU Allocation Ratio=1:12026.02 | 162.75 | — |