Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -62,6 +62,14 @@ exported both ways, whisper-tiny's decoder comes out at 198.0 MB exactly.
|
|
| 62 |
|
| 63 |
|
| 64 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
## Conversion
|
| 66 |
|
| 67 |
```bash
|
|
|
|
| 62 |
|
| 63 |
|
| 64 |
|
| 65 |
+
### Checked in the task's own units
|
| 66 |
+
|
| 67 |
+
Correlation is a first filter. These are the numbers that decide:
|
| 68 |
+
|
| 69 |
+
- **encoder int8** — measured end to end — word error rate against the fp32 encoder: mean WER 0.0% (worst clip 0.0%) over 5 spoken sentences, int8 encoder against the fp32 encoder with the same fp32 decoder and the same waveform; the fp32 arm transcribes all five correctly, so the comparison is against a working control rather than against noise.
|
| 70 |
+
|
| 71 |
+
The sensitivity of that test, measured by injecting random noise into whisper-tiny's encoder output: rel_l2 0.03 (what int8 actually costs) and 0.10 both give WER 0.000; 0.20 and 0.40 give 0.025. Five clean sentences leave headroom, so a pass means *does not break the transcript*, not *indistinguishable at any error level*.
|
| 72 |
+
|
| 73 |
## Conversion
|
| 74 |
|
| 75 |
```bash
|