twinlitenet — ExecuTorch XNNPACK
twinlitenet_xnnpack_fp32.pte (1.8 MB, fp32, XNNPACK-delegated)
- Source: chequanghuy/TwinLiteNet (pretrained/best.pth)
- License: MIT
- Input: [[1, 3, 360, 640]] — RGB 0-1, 360x640
- Output: drivable area [1,2,360,640] + lane line [1,2,360,640]
Verification (Mac arm64, executorch 1.4.0, torch 2.13.0)
Parity vs torch fp32 eager on random input:
| output | shape | max_abs_diff | corr |
|---|---|---|---|
| 0 | [1, 2, 360, 640] | 1.192e-05 | 1.000000 |
| 1 | [1, 2, 360, 640] | 1.049e-05 | 1.000000 |
Median latency over 10 runs (single Mac process, reference only — device numbers to follow): ExecuTorch 33.7 ms vs torch eager 23.3 ms.
Conversion
torch.export -> to_edge_transform_and_lower(XnnpackPartitioner) -> .pte (conversion scripts: executorch-models)
Notes: PReLU is excluded from XNNPACK delegation (XNNPACK PReLU segfaults at execute on macOS arm64 in executorch 1.4.0; minimal repro: a lone nn.PReLU(1)). PReLU runs on the portable kernel instead; outputs are bit-identical to the stock model.
Part of executorch-models — a verified .pte zoo for ExecuTorch. Conversion scripts and all models are indexed there.
- Downloads last month
- -