Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -29,6 +29,12 @@ All variants take and return fp32 tensors — swap the `.pte` file, keep your ap
|
|
| 29 |
\*Mac arm64, single process, median of 10 — a reference point for relative cost
|
| 30 |
only, not a device number (torch eager fp32 on the same machine: 85.0 ms).
|
| 31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
## Verification (executorch 1.4.0, torch 2.13.0)
|
| 33 |
|
| 34 |
Parity is measured against the fp32 eager model on random input; `corr` is
|
|
|
|
| 29 |
\*Mac arm64, single process, median of 10 — a reference point for relative cost
|
| 30 |
only, not a device number (torch eager fp32 on the same machine: 85.0 ms).
|
| 31 |
|
| 32 |
+
### Checked in the task's own units
|
| 33 |
+
|
| 34 |
+
Correlation is a first filter. These are the numbers that decide:
|
| 35 |
+
|
| 36 |
+
- **int8 (dynamic)** — delta-1 against the fp32 build is 0.994 median over ten real images and 0.975 at worst — for reference, depth models report delta-1 near 0.95 against ground truth, so the quantization error sits well inside the model's own error.
|
| 37 |
+
|
| 38 |
## Verification (executorch 1.4.0, torch 2.13.0)
|
| 39 |
|
| 40 |
Parity is measured against the fp32 eager model on random input; `corr` is
|