FAQ

Q: What Python versions are supported by the NPU compiler?
A: Currently, Python 2.7, Python 3.6, and Python 3.7 are supported.

Q: What TensorFlow versions are supported by the NPU compiler?
A: All versions between TensorFlow 1.4 and TensorFlow 1.15 are supported.

Q: What PyTorch versions are supported by the NPU compiler?
A: All versions between PyTorch 1.10 and PyTorch 1.13 are supported, but the Python version must be Python 3.7.

Q: What should I do if I encounter the error Some op types are not supported! when compiling a TensorFlow model?
A: First, check whether the model contains unnecessary OPs for deployment, such as RandomUniform, and remove them if present. Otherwise, consult GXchip engineers to see if the unsupported OP printed in the error can be added.

Q: What should I do if I encounter the error NPU Compiler don't support type xxx operator! when compiling a PyTorch model?
A: Consult GXchip engineers to see if the unsupported OP printed in the error can be added.

Q: Are dynamic RNN models supported?
A: No, they are not supported. The model structure needs to be modified to use a for loop instead.

Q: What should I do if the compilation fails with an unclear error message?
A: Update the NPU compiler to the latest version. If the issue persists, consult GXchip engineers.