| library_name: coremltools | |
| license: apache-2.0 | |
| tags: | |
| - coreml | |
| - document-layout-analysis | |
| - vision | |
| - macos | |
| base_model: PaddlePaddle/PP-DocLayoutV3_safetensors | |
| # PP-DocLayoutV3 Core ML | |
| FP16 Core ML conversion of | |
| [`PaddlePaddle/PP-DocLayoutV3_safetensors`](https://huggingface.co/PaddlePaddle/PP-DocLayoutV3_safetensors) | |
| for the [hugging-mac](https://github.com/hugging-mac/hugging-mac) SDK. | |
| The ML Program uses a fixed `pixel_values` input with shape `[1, 3, 800, 800]` | |
| and returns raw `logits`, `pred_boxes`, `order_logits`, and `out_masks` tensors. | |
| Preprocessing and document-layout postprocessing are implemented by the SDK. | |
| The package targets macOS 14 or newer. Conversion preserves the Apache-2.0 | |
| license of the source model. | |