PyTorch Mobile + NNAPI integration

yupri · January 21, 2021, 8:51am

Hi!
I’m looking into getting a YOLOv5 model running on NNAPI.
The doc ((Prototype) Convert MobileNetV2 to NNAPI — PyTorch Tutorials 1.7.1 documentation) states that the range of supported operators is limited.
Is it possible to get the listing of supported ops?
What is the strategy of the framework in a situation where the model uses both supported and unsupported ops? Does it bail out of NNAPI completely, or uses split execution, like TFLite?

IvanKobzarev · January 26, 2021, 5:37pm

The listing of supported ops you can find here:

github.com

pytorch/pytorch/blob/master/torch/backends/_nnapi/serializer.py#L585-L636


ADDER_MAP = {
    "prim::GetAttr": lambda self, node:
        self.add_getattr(node),
    "prim::Constant": lambda self, node:
        self.add_constant_node(node),
    "prim::ListConstruct": lambda self, node:
        self.add_list_construct(node),
    "prim::TupleConstruct": lambda self, node:
        self.add_tuple_construct(node),
    "aten::reshape": lambda self, node:
        self.add_reshape(node),
    "aten::quantize_per_tensor": lambda self, node:
        self.add_quantize(node),
    "aten::dequantize": lambda self, node:
        self.add_dequantize(node),
    "aten::add": lambda self, node:
        self.add_add_sub_op(node, NNAPI_OperationCode.ADD, NNAPI_FuseCode.FUSED_NONE),
    "aten::sub": lambda self, node:
        self.add_add_sub_op(node, NNAPI_OperationCode.SUB, NNAPI_FuseCode.FUSED_NONE),
    "aten::mul": lambda self, node:

This file has been truncated. show original

If operation is not supported - serialization will fail with exception pytorch/serializer.py at master · pytorch/pytorch · GitHub

Aby · April 21, 2022, 7:51pm

Hi what was your progress to get the Pytorch Yolov5 model to run on NNAPI?