external-cannbot-model-model-infer-fusion
Ascend NPU model inference fused operator optimization skill based on the PyTorch framework. Analyze model code, identify computation patterns that can be replaced with torch_npu fused operators, and generate replacement plans. Trigger scenarios: torch_npu fused operator replacement, inference operator adaptation for modules such as MoE/Attention/FFN/Norm, torch_npu API…