Use measured 1000-word speed benchmark
Browse files
README.md
CHANGED
|
@@ -98,10 +98,11 @@ on such inputs.
|
|
| 98 |
|
| 99 |
## Speed
|
| 100 |
|
| 101 |
-
**
|
| 102 |
-
loading. The int8 ONNX benchmark includes tokenization and rendering;
|
| 103 |
-
excludes download, speech recognition, and browser WASM. Speed varies by device
|
| 104 |
-
and input length. [Reproduce the benchmark](https://github.com/Rye-A1/ekko-danish-stt/blob/main/benchmarks/benchmark_pnc_latency.py)
|
|
|
|
| 105 |
|
| 106 |
## Files and use
|
| 107 |
|
|
|
|
| 98 |
|
| 99 |
## Speed
|
| 100 |
|
| 101 |
+
**1,000 Danish words in about 0.09 seconds** on an Apple M1 Pro CPU, after model
|
| 102 |
+
loading. The int8 ONNX benchmark includes tokenization, inference, and rendering;
|
| 103 |
+
it excludes download, speech recognition, and browser WASM. Speed varies by device
|
| 104 |
+
and input length. [Reproduce the benchmark](https://github.com/Rye-A1/ekko-danish-stt/blob/main/benchmarks/benchmark_pnc_latency.py)
|
| 105 |
+
with `--words 1000`.
|
| 106 |
|
| 107 |
## Files and use
|
| 108 |
|