LoSealL

Use this skill when exporting PyTorch models to ONNX using the modern torch.export (dynamo) path in HyperONNX. Best for models with dynamic control flow, complex tensor operations, and hierarchical custom operators defined via onnxscript.

Use this skill when working in HyperONNX and you need the legacy Torch ONNX export path with dynamo=False, especially for TorchScript-based export behavior, lower-version compatibility, custom attention registration, or post-export ONNX fixes.

Use when working on ONNXifier internals: ONNX graph rewrites, optimization passes, PassManager behavior, pass registration, CLI conversion flow, domain adapters, and pytest coverage for graph or pass changes. Triggers include add pass, debug pass, update ONNX transformation, fix failing pass test, inspect Rewriter or Pattern usage, and explain ONNXifier architecture.

Use when integrating with or calling the DDP (Delayed Dispatch Platform) backend API — submitting GPU jobs with ssh debug pods, editing pending jobs, checking status, fetching logs/outputs, admin cross-user management, or implementing auth flows. Covers all REST endpoints, session cookie auth, and the mock/k8s executor modes.

How to implement a custom inference Backend for the openai_http server and start the server with it. Use this skill whenever the user asks about creating or customizing backends, implementing inference logic, setting up run_server(), extending the openai_http server with custom model support, or working with BackendBase. Also trigger when the user wants to serve custom models via an OpenAI-compatible API endpoint.