LLM Inference on MindVLA LLM Component
0.1Latency (ms)M100
Evaluation results
| Method | Links | ||
|---|---|---|---|
| M100Inference Phase=decode, Hardware Platform=M100, Number of active clusters=122026.04 | 0.1 | 3 | |
| Thor-UInference Phase=decode, Hardware Platform=Thor-U2026.04 | 0.3 | — | |
| M100Inference Phase=prefill, Hardware Platform=M100, Number of active clusters=122026.04 | 0.84 | 2.1 | |
| Thor-UInference Phase=prefill, Hardware Platform=Thor-U2026.04 | 1.74 | — |