---

name: deployment-and-export description: "Export YOLOv7-d2 models to ONNX or TorchScript, inspect ONNXRuntime inputs, convert DETR checkpoints, and reason about TensorRT or quantization deployment." metadata: disco-role: operating disable-model-invocation: true license: GPL 3.0

Deployment and Export

Use this sub-skill when the user asks about YOLOv7-d2 ONNX/TorchScript export, ONNXRuntime inference, model IO inspection, DETR/AnchorDETR/SMCA checkpoint conversion, TensorRT, or quantization.

Start here

1. Confirm the user has a trained checkpoint, matching config, and sample input image before export. Read references/export-onnx.md.
2. Build an export command with scripts/build_export_command.py.
3. Inspect exported ONNX files with scripts/inspect_onnx_model.py, then read references/onnxruntime-inference.md.
4. For reference DETR checkpoints, use scripts/convert_detr_checkpoint.py and read references/checkpoint-conversion.md.
5. For TensorRT or quantization, read references/quantization-and-tensorrt.md before installing toolchains.
6. Use references/troubleshooting.md for export/import/operator/runtime errors.

Important distinctions

• Export is a PyTorch/Detectron2 model operation that needs the original model code, config, weights, and sample input shape.
• ONNXRuntime inference needs a completed ONNX file and postprocessing assumptions matching the exported model family.
• TensorRT and quantization are optional hardware/toolchain workflows, not automatically verified by a successful ONNX export.

Boundaries

• For training configs and checkpoints before export, read ../training-and-configuration/SKILL.md.
• For ordinary PyTorch checkpoint demo/evaluation, read ../inference-and-evaluation/SKILL.md.