kernels-benchmarks
/
moe_benchmarks
/megablocks_yamoe
/artifacts
/megablocks_run
/megablocks_results.json
| { | |
| "implementation": "megablocks_results", | |
| "config": { | |
| "warmup": 10, | |
| "iters": 50, | |
| "device": "cuda", | |
| "dtype": "torch.float32", | |
| "tokens": 100, | |
| "vary_inputs": true | |
| }, | |
| "stats": { | |
| "avg_ms": 3.8653191399976095, | |
| "min_ms": 0.8422269999073251, | |
| "max_ms": 8.544625000013184, | |
| "std_ms": 3.690530253469649, | |
| "p50_ms": 0.88288749992671, | |
| "p95_ms": 8.536876499982782, | |
| "p99_ms": 8.542453809967583, | |
| "num_iters": 50, | |
| "tokens_per_s": 25871.084994048342, | |
| "throughput_variance": 50674.824252369 | |
| }, | |
| "output_sum": 6.4738850593566895 | |
| } |