I'm not sure what the intent of the Async variations of benchmarks are.
For example, here's one result on my M1 Pro:
| Feature Extraction on webgpu |
957.76 |
142.965 |
| Feature Extraction on webgpu/Benchmark |
957.76 |
142.965 |
| Feature Extraction on webgpu/Benchmark/Sync |
957.745 |
142.955 |
| Feature Extraction on webgpu/Benchmark/Async |
0.01499998569 |
0.009999990463 |
The async variations seem to only measure the launch of the work, and don't wait for the result to resolve. Is that the intention?