Train a model with the TVM


Hi there,

I have a question about model training with the TVM.

I want to add my NPU accelerator to the TVM. Then I want to use the TVM to train a model with my NPU.

I’m new at the DP/ML and do not fully understand

1. can I import a model from a framework (for example TensorFlow: ) and use the TVM to train the model (calculate the optimal weights) ??? Maybe do you have any example for it ?

I have read

So my question is the same as the first one link.

Also I have read the v0.6 roadmap and noticed the following features:


  • First-order auto differentiation
  • Gradient operators
  • High-order auto differentiation

2. Will the TVM v0.6 fully support the training ?? Or it will only partial support?

Best Regards,