quark-onnx-debug

SkillDev tools

Diagnose failed Quark ONNX installation, calibration, quantization, custom-op compilation, or export attempts. Use when the user reports an error, stack trace, invalid artifact, missing dependency, ORT execution-provider mismatch, silent CPU fallback, OOM during calibration, custom-op load failure (BFPQuantizeDequantize / MXQuantizeDequantize / Extended*), or unexpected quantization results from the ONNX flow. Trigger for "Quark ONNX error", "onnxruntime error", "quantize_static failed", "calibration crashed", "CUDAExecutionProvider not available", "ROCMExecutionProvider not available", "custom op library load failed", "model.onnx larger than 2GB", "external data not found", "AdaRound diverged", "GPTQ ONNX failed", "QuaRot failed", "NPU power-of-2 scale", any Python traceback mentioning quark.onnx / onnxruntime / onnx.

Available today. Use it from your connected AI after setup.

No other account needed.

Then ask your AI: use the quark-onnx-debug skill