How to use adaround when the model requires multiple inputs

[unique-null]

when I use adaround as below:

the error appeares as follows:

the env shows:

how can I fix the error. thx.

[quic-hitameht]

When configuring Adaround parameters, along with data_loader, forward_fn is required which is simply an adapter function that performs forward pass given a model and model inputs yielded from the provided data loader. API doc for Adaround: https://quic.github.io/aimet-pages/releases/latest/api_docs/torch_adaround.html#api-torch-adaround

By default, the assumption is that the provided data loader is labeled and it only passes the first element to the model if data loader yields the tuple and this restriction is I guess not very well documented and needs to be fixed. Here is the definition of default forward function: https://github.com/quic/aimet/blob/develop/TrainingExtensions/torch/src/python/aimet_torch/utils.py#L170

All you need to do is define your own custom forward_fn and pass it when creating AdaroundParameters object:

def forward_fn(model, inputs):
    model(*inputs)

Let us know if you have any other questions.