File size: 2,474 Bytes
cbe17fb 6d663c2 cbe17fb 03036d3 6d663c2 03036d3 6d663c2 a91d938 03036d3 6d663c2 cbe17fb 6d663c2 03036d3 cbe17fb 03036d3 cbe17fb 03036d3 cbe17fb 6d663c2 ce28719 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 | ---
license: bsd-3-clause
tags:
- executorch
- xnnpack
- pte
- on-device
- image-classification
---
# efficientnet_b1 — ExecuTorch
- **Source**: torchvision efficientnet_b1 IMAGENET1K_V2
- **License**: BSD-3-Clause
- **Input**: [[1, 3, 240, 240]] — RGB, ImageNet norm, 240x240
- **Output**: ImageNet logits [1,1000]
## Variants
All variants take and return fp32 tensors — swap the `.pte` file, keep your app code.
| build | file | size (MB) | parity vs fp32 eager (worst corr) | Mac median (ms)* |
|-----------|------|-----------|------------------------------------|------------------|
| fp32 | `efficientnet_b1_xnnpack_fp32.pte` | 31.2 | 1.000000 | 9.3 |
| fp16 | `efficientnet_b1_xnnpack_fp16.pte` | 28.8 | 0.999816 | 54.9 |
| Core ML (fp16, iOS) | `efficientnet_b1_coreml_all.pte` | 16.3 | 0.992817 — see below | 0.6 |
The Core ML build is the same graph lowered to Apple's Neural Engine instead of
XNNPACK, which is CPU-only. Measured on an iPhone 17 Pro across seven models, it
runs **3.5x to 13.9x faster (median 12x)** at roughly half the file size — for
example Depth-Anything-V2-Small at 500.8 ms against 42.7 ms, and MODNet at 81.7 ms
against 5.9 ms. It computes in fp16 and is iOS-only; the XNNPACK files stay the
portable option and are what runs on Android.
\*Mac arm64, single process, median of 10 — a reference point for relative cost
only, not a device number (torch eager fp32 on the same machine: 352.7 ms).
### Checked in the task's own units
Correlation is a first filter. These are the numbers that decide:
- **Core ML (fp16, iOS)** — measured in the units that matter for this model — fraction of images keeping the fp32 top-1 label: 9 of 10 images keep the fp32 top-1 label.
### Builds that did not earn a slot
- **int8 is not shipped**: measured in the units that matter for this model — fraction of images keeping the fp32 top-1 label: 0 of 10 images keep the fp32 top-1 label.
## Verification (executorch 1.4.0, torch 2.13.0)
Parity is measured against the fp32 eager model on real image input; `corr` is
the correlation over all elements of each output tensor.
| output | shape | max_abs_diff | corr |
|--------|-------|--------------|------|
| 0 | [1, 1000] | 1.490e-05 | 1.000000 |
XNNPACK delegate coverage (fp32): 100.0% (410/410 ops)
## Conversion
torch.export -> to_edge_transform_and_lower(partitioner) -> .pte
(conversion scripts: [executorch-models](https://github.com/john-rocky/executorch-models))
|