# Training Throughput Measurement on ManiSkill Source: [Training Throughput Measurement on ManiSkill]() Collection index: [llms.txt]() Data updated: 2026-02-26T06:07:34.981Z Evaluation of the training throughput performance (e.g., samples processed per second) when training models using the ManiSkill dataset. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Training Throughput Measurement | | Dataset (including any stated split) | ManiSkill | | Metric | Training Throughput | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 15 of 15 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Training Throughput](): Higher is better; unit: SCALAR. - [Throughput Increase (%)](): Higher is better; unit: PERCENT. ## Recorded results | Method | Training Throughput | Result date | Report paper | | --- | --- | --- | --- | | Train Async | 436.32 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Colocated | 370.26 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Streamer | 280.07 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Rollout Async | 275.36 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Disaggregated | 257.21 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Train Async | 244.61 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Colocated | 232.23 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Disaggregated | 150.59 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Streamer | 141.49 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Rollout Async | 139.32 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Colocated | 132.56 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Train Async | 126.65 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Streamer | 71.61 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Rollout Async | 69.07 | Not recorded | 698575cfc97a0b3e3b5d5e1a | | Disaggregated | 60.64 | Not recorded | 698575cfc97a0b3e3b5d5e1a | ### Train Async Training Throughput: 436.32. Result ID: 6996018c5698665ec437d701. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"32 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Colocated Training Throughput: 370.26. Result ID: 6996018c5698665ec437d6ff. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"32 GPUs"}. This record does not assert independent reproduction. ### Streamer Training Throughput: 280.07. Result ID: 6996018c5698665ec437d703. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"32 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Rollout Async Training Throughput: 275.36. Result ID: 6996018c5698665ec437d702. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"32 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Disaggregated Training Throughput: 257.21. Result ID: 6996018c5698665ec437d700. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"32 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Train Async Training Throughput: 244.61. Result ID: 6996018c5698665ec437d6fc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"16 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Colocated Training Throughput: 232.23. Result ID: 6996018c5698665ec437d6fa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"16 GPUs"}. This record does not assert independent reproduction. ### Disaggregated Training Throughput: 150.59. Result ID: 6996018c5698665ec437d6fb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"16 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Streamer Training Throughput: 141.49. Result ID: 6996018c5698665ec437d6fe. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"16 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Rollout Async Training Throughput: 139.32. Result ID: 6996018c5698665ec437d6fd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"16 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Colocated Training Throughput: 132.56. Result ID: 6996018c5698665ec437d6f5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"8 GPUs"}. This record does not assert independent reproduction. ### Train Async Training Throughput: 126.65. Result ID: 6996018c5698665ec437d6f7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"8 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Streamer Training Throughput: 71.61. Result ID: 6996018c5698665ec437d6f9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"8 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Rollout Async Training Throughput: 69.07. Result ID: 6996018c5698665ec437d6f8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"8 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. ### Disaggregated Training Throughput: 60.64. Result ID: 6996018c5698665ec437d6f6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"VLA Backbone":"pi0","Number of GPUs":"8 GPUs","GPU Allocation Ratio":"1:1"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.