| license: apache-2.0 | |
| tags: | |
| - pp-doclayoutv3 | |
| - tensorrt | |
| - onnx | |
| - document-layout-analysis | |
| pipeline_tag: object-detection | |
| library_name: paddlex | |
| # PP-DocLayoutV3 TensorRT | |
| ## Contents | |
| ```text | |
| onnx/inference.onnx | |
| paddlex/inference.json | |
| paddlex/inference.yml | |
| engines/sm89-trt10.15-cuda12.6-fp16-b1_8/pp_doclayout_v3.engine | |
| engines/sm89-trt10.15-cuda12.6-fp16-b1_8/engine_manifest.json | |
| artifact_manifest.json | |
| scripts/build_engine.sh | |
| ``` | |
| ## Engine | |
| ```text | |
| GPU: RTX 4090 / sm89 | |
| TensorRT: 10.15.1.29 | |
| CUDA: 12.6 | |
| Precision: FP16 | |
| Profile: batch 1-8, image 3x800x800 | |
| ``` | |
| TensorRT engines are runtime-specific. Rebuild from `onnx/inference.onnx` if the engine is not compatible with the deployment GPU or TensorRT version. | |
| ## Rebuild | |
| ```bash | |
| bash scripts/build_engine.sh | |
| ``` | |