What problem does it solve?
Debugging ExecuTorch deployment issues often requires digging through thousands of scattered GitHub issues and discussions. This Skill provides a synthesized tribal knowledge base covering QNN, XNNPACK, Vulkan, CoreML, Arm, and Cadence backends so you can find proven answers to export, quantization, and runtime problems quickly.
Core Features & Use Cases
- Backend-Specific Knowledge: Query curated articles on QNN/HTP SoC compatibility, XNNPACK CPU delegation, Vulkan shader bugs, CoreML behavior, Arm Ethos-U, and Cadence Xtensa.
- Quantization & Export Guidance: Find quantization recipes, accuracy debugging steps, and torch.export pitfalls for LLM and vision models.
- Troubleshooting with Citations: Get answers for build failures, runtime crashes, and performance issues with source issue citations and confidence flags for single-source workarounds.
- Use Case: You hit an HTP error when delegating a quantized model to a Snapdragon V68 SoC. Query the knowledge base to find the known cause, the reported workaround, and whether to verify against the official QNN op support documentation.
Quick Start
Ask the executorch-kb skill a question such as why your QNN delegation fails on a V68 HTP or which quantization recipe to use for a Llama model.
