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: 375.5 ms).
|
| 31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
### Precisions that did not earn a slot
|
| 33 |
|
| 34 |
- **fp16 is not shipped**: it comes out at 100% of the fp32 file (176.1 MB vs 176.1 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.
|
|
|
|
| 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: 375.5 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** — measured in the units that matter for this model: mask IoU at 0.5, median 0.9994 over five real images (worst 0.9916) against the fp32 build.
|
| 37 |
+
|
| 38 |
### Precisions that did not earn a slot
|
| 39 |
|
| 40 |
- **fp16 is not shipped**: it comes out at 100% of the fp32 file (176.1 MB vs 176.1 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.
|