--- license: agpl-3.0 pipeline_tag: object-detection library_name: executorch --- # yolo26 This repository hosts the **yolo26** models exported for the [React Native ExecuTorch](https://www.npmjs.com/package/react-native-executorch) library as ExecuTorch `.pte` programs, ready to run on device. Upstream model: [Yolo26](https://docs.ultralytics.com/tasks/detect/) ## Variants | Path | Backend | Precision | | --- | --- | --- | | `l/coreml/yolo26l_384_coreml_fp16.pte` | coreml | fp16 | | `l/coreml/yolo26l_512_coreml_fp16.pte` | coreml | fp16 | | `l/coreml/yolo26l_640_coreml_fp16.pte` | coreml | fp16 | | `l/vulkan/yolo26l_384_vulkan_fp16.pte` | vulkan | fp16 | | `l/vulkan/yolo26l_512_vulkan_fp16.pte` | vulkan | fp16 | | `l/vulkan/yolo26l_640_vulkan_fp16.pte` | vulkan | fp16 | | `l/xnnpack/yolo26l_384_xnnpack_fp32.pte` | xnnpack | fp32 | | `l/xnnpack/yolo26l_384_xnnpack_int8.pte` | xnnpack | int8 | | `l/xnnpack/yolo26l_512_xnnpack_fp32.pte` | xnnpack | fp32 | | `l/xnnpack/yolo26l_512_xnnpack_int8.pte` | xnnpack | int8 | | `l/xnnpack/yolo26l_640_xnnpack_fp32.pte` | xnnpack | fp32 | | `l/xnnpack/yolo26l_640_xnnpack_int8.pte` | xnnpack | int8 | | `m/coreml/yolo26m_384_coreml_fp16.pte` | coreml | fp16 | | `m/coreml/yolo26m_512_coreml_fp16.pte` | coreml | fp16 | | `m/coreml/yolo26m_640_coreml_fp16.pte` | coreml | fp16 | | `m/vulkan/yolo26m_384_vulkan_fp16.pte` | vulkan | fp16 | | `m/vulkan/yolo26m_512_vulkan_fp16.pte` | vulkan | fp16 | | `m/vulkan/yolo26m_640_vulkan_fp16.pte` | vulkan | fp16 | | `m/xnnpack/yolo26m_384_xnnpack_fp32.pte` | xnnpack | fp32 | | `m/xnnpack/yolo26m_384_xnnpack_int8.pte` | xnnpack | int8 | | `m/xnnpack/yolo26m_512_xnnpack_fp32.pte` | xnnpack | fp32 | | `m/xnnpack/yolo26m_512_xnnpack_int8.pte` | xnnpack | int8 | | `m/xnnpack/yolo26m_640_xnnpack_fp32.pte` | xnnpack | fp32 | | `m/xnnpack/yolo26m_640_xnnpack_int8.pte` | xnnpack | int8 | | `n/coreml/yolo26n_384_coreml_fp16.pte` | coreml | fp16 | | `n/coreml/yolo26n_512_coreml_fp16.pte` | coreml | fp16 | | `n/coreml/yolo26n_640_coreml_fp16.pte` | coreml | fp16 | | `n/vulkan/yolo26n_384_vulkan_fp16.pte` | vulkan | fp16 | | `n/vulkan/yolo26n_512_vulkan_fp16.pte` | vulkan | fp16 | | `n/vulkan/yolo26n_640_vulkan_fp16.pte` | vulkan | fp16 | | `n/xnnpack/yolo26n_384_xnnpack_fp32.pte` | xnnpack | fp32 | | `n/xnnpack/yolo26n_384_xnnpack_int8.pte` | xnnpack | int8 | | `n/xnnpack/yolo26n_512_xnnpack_fp32.pte` | xnnpack | fp32 | | `n/xnnpack/yolo26n_512_xnnpack_int8.pte` | xnnpack | int8 | | `n/xnnpack/yolo26n_640_xnnpack_fp32.pte` | xnnpack | fp32 | | `n/xnnpack/yolo26n_640_xnnpack_int8.pte` | xnnpack | int8 | | `s/coreml/yolo26s_384_coreml_fp16.pte` | coreml | fp16 | | `s/coreml/yolo26s_512_coreml_fp16.pte` | coreml | fp16 | | `s/coreml/yolo26s_640_coreml_fp16.pte` | coreml | fp16 | | `s/vulkan/yolo26s_384_vulkan_fp16.pte` | vulkan | fp16 | | `s/vulkan/yolo26s_512_vulkan_fp16.pte` | vulkan | fp16 | | `s/vulkan/yolo26s_640_vulkan_fp16.pte` | vulkan | fp16 | | `s/xnnpack/yolo26s_384_xnnpack_fp32.pte` | xnnpack | fp32 | | `s/xnnpack/yolo26s_384_xnnpack_int8.pte` | xnnpack | int8 | | `s/xnnpack/yolo26s_512_xnnpack_fp32.pte` | xnnpack | fp32 | | `s/xnnpack/yolo26s_512_xnnpack_int8.pte` | xnnpack | int8 | | `s/xnnpack/yolo26s_640_xnnpack_fp32.pte` | xnnpack | fp32 | | `s/xnnpack/yolo26s_640_xnnpack_int8.pte` | xnnpack | int8 | | `x/coreml/yolo26x_384_coreml_fp16.pte` | coreml | fp16 | | `x/coreml/yolo26x_512_coreml_fp16.pte` | coreml | fp16 | | `x/coreml/yolo26x_640_coreml_fp16.pte` | coreml | fp16 | | `x/vulkan/yolo26x_384_vulkan_fp16.pte` | vulkan | fp16 | | `x/vulkan/yolo26x_512_vulkan_fp16.pte` | vulkan | fp16 | | `x/vulkan/yolo26x_640_vulkan_fp16.pte` | vulkan | fp16 | | `x/xnnpack/yolo26x_384_xnnpack_fp32.pte` | xnnpack | fp32 | | `x/xnnpack/yolo26x_384_xnnpack_int8.pte` | xnnpack | int8 | | `x/xnnpack/yolo26x_512_xnnpack_fp32.pte` | xnnpack | fp32 | | `x/xnnpack/yolo26x_512_xnnpack_int8.pte` | xnnpack | int8 | | `x/xnnpack/yolo26x_640_xnnpack_fp32.pte` | xnnpack | fp32 | | `x/xnnpack/yolo26x_640_xnnpack_int8.pte` | xnnpack | int8 | ## XNNPACK int8 The `xnnpack/*_int8.pte` files are static int8 (PT2E, per-channel weights) with the Detect head and the first convolution kept fp32. COCO mAP50-95 on the first 500 val2017 images against the fp32 file at the same resolution: | | 384 fp32 | 384 int8 | 512 fp32 | 512 int8 | 640 fp32 | 640 int8 | | --- | --- | --- | --- | --- | --- | --- | | n | 36.91 | 36.83 | 41.24 | 41.03 | 44.58 | 43.92 | | s | 45.40 | 45.03 | 48.44 | 47.91 | 50.67 | 50.06 | | m | 51.51 | 50.73 | 54.19 | 53.75 | 56.30 | 56.17 | | l | 53.26 | 52.81 | 56.18 | 56.04 | 58.10 | 58.24 | | x | 55.96 | 55.78 | 58.83 | 58.71 | 59.94 | 59.94 | ## Unshipped backends A backend directory carrying a `NOTES.md` and no `.pte` was exported and not published. Its note records what went wrong. - `l/mlx` ([why](https://huggingface.co/software-mansion/react-native-executorch-yolo26/blob/main/l/mlx/NOTES.md)) - `m/mlx` ([why](https://huggingface.co/software-mansion/react-native-executorch-yolo26/blob/main/m/mlx/NOTES.md)) - `n/mlx` ([why](https://huggingface.co/software-mansion/react-native-executorch-yolo26/blob/main/n/mlx/NOTES.md)) - `s/mlx` ([why](https://huggingface.co/software-mansion/react-native-executorch-yolo26/blob/main/s/mlx/NOTES.md)) - `x/mlx` ([why](https://huggingface.co/software-mansion/react-native-executorch-yolo26/blob/main/x/mlx/NOTES.md)) ## Repository structure ``` config.json 28 B l/coreml/config.json 2.6 kB l/coreml/yolo26l_384_coreml_fp16.pte 48.5 MB l/coreml/yolo26l_512_coreml_fp16.pte 48.5 MB l/coreml/yolo26l_640_coreml_fp16.pte 48.5 MB l/mlx/NOTES.md 2.5 kB l/vulkan/config.json 2.7 kB l/vulkan/yolo26l_384_vulkan_fp16.pte 47.6 MB l/vulkan/yolo26l_512_vulkan_fp16.pte 47.6 MB l/vulkan/yolo26l_640_vulkan_fp16.pte 47.6 MB l/xnnpack/config.json 4.9 kB l/xnnpack/yolo26l_384_xnnpack_fp32.pte 94.9 MB l/xnnpack/yolo26l_384_xnnpack_int8.pte 28.4 MB l/xnnpack/yolo26l_512_xnnpack_fp32.pte 94.9 MB l/xnnpack/yolo26l_512_xnnpack_int8.pte 28.4 MB l/xnnpack/yolo26l_640_xnnpack_fp32.pte 95.0 MB l/xnnpack/yolo26l_640_xnnpack_int8.pte 28.4 MB m/coreml/config.json 2.6 kB m/coreml/yolo26m_384_coreml_fp16.pte 39.8 MB m/coreml/yolo26m_512_coreml_fp16.pte 39.8 MB m/coreml/yolo26m_640_coreml_fp16.pte 39.8 MB m/mlx/NOTES.md 2.5 kB m/vulkan/config.json 2.7 kB m/vulkan/yolo26m_384_vulkan_fp16.pte 39.1 MB m/vulkan/yolo26m_512_vulkan_fp16.pte 39.1 MB m/vulkan/yolo26m_640_vulkan_fp16.pte 39.1 MB m/xnnpack/config.json 4.9 kB m/xnnpack/yolo26m_384_xnnpack_fp32.pte 78.1 MB m/xnnpack/yolo26m_384_xnnpack_int8.pte 24.1 MB m/xnnpack/yolo26m_512_xnnpack_fp32.pte 78.1 MB m/xnnpack/yolo26m_512_xnnpack_int8.pte 24.1 MB m/xnnpack/yolo26m_640_xnnpack_fp32.pte 78.1 MB m/xnnpack/yolo26m_640_xnnpack_int8.pte 24.1 MB n/coreml/config.json 2.6 kB n/coreml/yolo26n_384_coreml_fp16.pte 5.4 MB n/coreml/yolo26n_512_coreml_fp16.pte 5.4 MB n/coreml/yolo26n_640_coreml_fp16.pte 5.4 MB n/mlx/NOTES.md 2.5 kB n/vulkan/config.json 2.7 kB n/vulkan/yolo26n_384_vulkan_fp16.pte 4.8 MB n/vulkan/yolo26n_512_vulkan_fp16.pte 4.8 MB n/vulkan/yolo26n_640_vulkan_fp16.pte 4.8 MB n/xnnpack/config.json 4.9 kB n/xnnpack/yolo26n_384_xnnpack_fp32.pte 9.4 MB n/xnnpack/yolo26n_384_xnnpack_int8.pte 3.0 MB n/xnnpack/yolo26n_512_xnnpack_fp32.pte 9.4 MB n/xnnpack/yolo26n_512_xnnpack_int8.pte 3.0 MB n/xnnpack/yolo26n_640_xnnpack_fp32.pte 9.4 MB n/xnnpack/yolo26n_640_xnnpack_int8.pte 3.0 MB s/coreml/config.json 2.6 kB s/coreml/yolo26s_384_coreml_fp16.pte 18.9 MB s/coreml/yolo26s_512_coreml_fp16.pte 18.9 MB s/coreml/yolo26s_640_coreml_fp16.pte 18.9 MB s/mlx/NOTES.md 2.5 kB s/vulkan/config.json 2.7 kB s/vulkan/yolo26s_384_vulkan_fp16.pte 18.3 MB s/vulkan/yolo26s_512_vulkan_fp16.pte 18.3 MB s/vulkan/yolo26s_640_vulkan_fp16.pte 18.3 MB s/xnnpack/config.json 4.9 kB s/xnnpack/yolo26s_384_xnnpack_fp32.pte 36.4 MB s/xnnpack/yolo26s_384_xnnpack_int8.pte 10.8 MB s/xnnpack/yolo26s_512_xnnpack_fp32.pte 36.4 MB s/xnnpack/yolo26s_512_xnnpack_int8.pte 10.8 MB s/xnnpack/yolo26s_640_xnnpack_fp32.pte 36.5 MB s/xnnpack/yolo26s_640_xnnpack_int8.pte 10.8 MB x/coreml/config.json 2.6 kB x/coreml/yolo26x_384_coreml_fp16.pte 108 MB x/coreml/yolo26x_512_coreml_fp16.pte 108 MB x/coreml/yolo26x_640_coreml_fp16.pte 108 MB x/mlx/NOTES.md 2.5 kB x/vulkan/config.json 2.7 kB x/vulkan/yolo26x_384_vulkan_fp16.pte 107 MB x/vulkan/yolo26x_512_vulkan_fp16.pte 107 MB x/vulkan/yolo26x_640_vulkan_fp16.pte 107 MB x/xnnpack/config.json 4.9 kB x/xnnpack/yolo26x_384_xnnpack_fp32.pte 213 MB x/xnnpack/yolo26x_384_xnnpack_int8.pte 63.1 MB x/xnnpack/yolo26x_512_xnnpack_fp32.pte 213 MB x/xnnpack/yolo26x_512_xnnpack_int8.pte 63.1 MB x/xnnpack/yolo26x_640_xnnpack_fp32.pte 213 MB x/xnnpack/yolo26x_640_xnnpack_int8.pte 63.1 MB ``` ## Compatibility These files are published for the **ExecuTorch v1.4.1** runtime. ExecuTorch gives no forward compatibility guarantee, so an older runtime may fail to load them. To use them in React Native ExecuTorch, pass the model constant shipped in the library's model registry to the corresponding task pipeline. See the [documentation](https://docs.swmansion.com/react-native-executorch/docs/fundamentals/downloading-models). To load these files in your own ExecuTorch runtime, read the [compatibility note](https://github.com/pytorch/executorch/blob/main/runtime/COMPATIBILITY.md) first.