Transformers.js - WebGPU Benchmark Results
#5
by
pcuenq
HF staff
- opened
| Batch Size | WASM (ms) | WebGPU (ms) |
| 1 | 531.60 | 15.20 |
| 2 | 1131.50 | 29.50 |
| 4 | 2257.20 | 48.20 |
| 8 | 4335.20 | 81.80 |
| 16 | 8963.40 | 146.50 |
| 32 | 17338.80 | 284.80 |
- Model: Xenova/all-MiniLM-L6-v2
- Quantized: false
- Sequence length: 512
- Browser: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/122.0.0.0 Safari/537.36
Something strange happened here, the times for bs=64 and bs=128 seem to correspond to sizes 1 and 2.
(I deleted the last 2 rows as it was probably due to an interrupt that continued to run)