Bug report when saving QuantizationSimModel encoding files for LSTM module

[XanxusCrypto]

This is a related issue with Bug report when using QuantizationSimModel to produce the quantization model for LSTM module.

After several PTQ and QAT of the QuantizationSimModel, I used the export API to export the onnx model and encoding files.
While I got an error for the following code in _aimet_torch/base/quantsim.py:

        for name in op_names:
            if name in end_op_names:
                output_tensors.extend(op_to_io_tensor_map[name].outputs)
            else:
                intermediate_tensors.extend(op_to_io_tensor_map[name].outputs)

intermediate_tensors.extend(op_to_io_tensor_map[name].outputs)
AttributeError: 'list' object has no attribute 'outputs'

Because the intermediate output of LSTM module was list. I tried to print them and got:

        print(name)
        if isinstance(op_to_io_tensor_map[name], list):
            print(op_to_io_tensor_map[name])

rnn_sequence_block.encoder
rnn_sequence_block.encoder#1
rnn_sequence_block.encoder#2
rnn_sequence_block.encoder#3
rnn_sequence_block.encoder#2-1
rnn_sequence_block.encoder#3-1
rnn_sequence_block.encoder#4
rnn_sequence_block.encoder#4-1
rnn_sequence_block.encoder#0-1
rnn_sequence_block.encoder#6
rnn_sequence_block.encoder#7-1
rnn_sequence_block.encoder#root_node
[<aimet_torch.defs.OpToIOTensors object at 0x7f0e50c58250>, <aimet_torch.defs.OpToIOTensors object at 0x7f0e50c2b7c0>]

intermediate_tensors.extend(op_to_io_tensor_map[name].outputs)
AttributeError: 'list' object has no attribute 'outputs'

Thus I tried to work around the bug:

   for name in op_names:
            if name in end_op_names:
                output_tensors.extend(op_to_io_tensor_map[name].outputs)
            else:
                if isinstance(op_to_io_tensor_map[name], list):
                    print(f"Name list f{name}")
                    print(op_to_io_tensor_map[name])
                    for named_tensor in op_to_io_tensor_map[name]:
                        intermediate_tensors.extend(named_tensor.outputs)
                else:
                    print(f"Name f{name}")
                    print(op_to_io_tensor_map[name])
                    intermediate_tensors.extend(op_to_io_tensor_map[name].outputs)

and still got an error, which indicated that the

Param tensor {weight_ih_lo, weight_hh_l0, bias_ih_l0, bias__hh_l0} not found in the valid param set.

Hope to solve this soon!

BRs.