PT to ONNX Converter

This tool provides an online workflow to convert PyTorch models saved in .pt or .pth into ONNX format, enabling deployment across runtime environments such as ONNX Runtime, TensorRT, or other inference backends. It targets ML engineers, AI researchers, and deployment engineers who need framework interoperability and optimized inference paths. The converter focuses on preserving model structure, weights, and input/output semantics during export. Core logic loads the PT model, optionally scripts or traces the model, and calls torch.onnx.export with user-specified options. It supports selecting an ONNX opset_version, dynamic_axes, and explicit input_names/output_names. After export, it runs a lightweight validation to ensure the ONNX graph is well-formed and that basic shape/type consistency holds. When unsupported operators are encountered, the tool reports diagnostics and suggests adjustments or an alternative export path. Audience benefits include faster cross-framework deployment, easier experimentation with backend accelerators, and a centralized workflow for validation. The system emphasizes transparency in export decisions and provides clear failure reasons, enabling iterative improvements. Core features include: PT input support (.pt/.pth), optional scripting or tracing, configurable opset_version, dynamic_axes, input/output naming, post-export validation, and a lightweight compatibility report. Use cases cover production deployment, model prototyping across platforms, and edge deployment prep.