Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -21,7 +21,6 @@ All variants take and return fp32 tensors — swap the `.pte` file, keep your ap
|
|
| 21 |
| precision | file | size (MB) | parity vs fp32 eager (worst corr) | Mac median (ms)* |
|
| 22 |
|-----------|------|-----------|------------------------------------|------------------|
|
| 23 |
| fp32 | `u2net_xnnpack_fp32.pte` | 176.0 | 1.000000 | 56.5 |
|
| 24 |
-
| int8 | `u2net_xnnpack_int8.pte` | 44.3 | 0.980186 | 33.0 |
|
| 25 |
|
| 26 |
\*Mac arm64, single process, median of 10 — a reference point for relative cost
|
| 27 |
only, not a device number (torch eager fp32 on the same machine: 138.5 ms).
|
|
@@ -29,6 +28,7 @@ only, not a device number (torch eager fp32 on the same machine: 138.5 ms).
|
|
| 29 |
### Precisions that did not earn a slot
|
| 30 |
|
| 31 |
- **fp16 is not shipped**: it comes out at 100% of the fp32 file (176.0 MB vs 176.0 MB), so it buys nothing. XNNPACK serializes convolution weights as fp32 no matter what dtype the graph carries, so on a conv-heavy model fp16 saves no disk and only adds cast operations. Reach for int8 here, not fp16.
|
|
|
|
| 32 |
|
| 33 |
## Verification (executorch 1.4.0, torch 2.13.0)
|
| 34 |
|
|
|
|
| 21 |
| precision | file | size (MB) | parity vs fp32 eager (worst corr) | Mac median (ms)* |
|
| 22 |
|-----------|------|-----------|------------------------------------|------------------|
|
| 23 |
| fp32 | `u2net_xnnpack_fp32.pte` | 176.0 | 1.000000 | 56.5 |
|
|
|
|
| 24 |
|
| 25 |
\*Mac arm64, single process, median of 10 — a reference point for relative cost
|
| 26 |
only, not a device number (torch eager fp32 on the same machine: 138.5 ms).
|
|
|
|
| 28 |
### Precisions that did not earn a slot
|
| 29 |
|
| 30 |
- **fp16 is not shipped**: it comes out at 100% of the fp32 file (176.0 MB vs 176.0 MB), so it buys nothing. XNNPACK serializes convolution weights as fp32 no matter what dtype the graph carries, so on a conv-heavy model fp16 saves no disk and only adds cast operations. Reach for int8 here, not fp16.
|
| 31 |
+
- **int8 is not shipped**: mask IoU 0.814 median and 0.212 at worst against fp32. The int8 build systematically shrinks weak saliency: on a photo where fp32 marks 2.2% of pixels foreground it marks 0.5%. On images with an unambiguous subject it agrees closely (0.99), but that is not a promise this model can make.
|
| 32 |
|
| 33 |
## Verification (executorch 1.4.0, torch 2.13.0)
|
| 34 |
|