coreml转换器在numpy 1.20上会报错

[Gemfield]

coreml转换器在numpy 1.20上会报错

[__MHGL__]

coreml转换问题

numpy版本问题

• numpy == 1.21.0
• coremltools == 0.4.1
• error

File "/home/liyang/HomePod/deepvac/deepvac/cast/coreml.py", line 60, in process
    coreml_model = coremltools.convert(model=model, inputs=[input], 
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/_converters_entry.py", line 175, in convert
    mlmodel = mil_convert(
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/converter.py", line 128, in mil_convert
    proto = mil_convert_to_proto(model, convert_from, convert_to,
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/converter.py", line 171, in mil_convert_to_proto
    prog = frontend_converter(model, **kwargs)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/converter.py", line 85, in __call__
    return load(*args, **kwargs)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 83, in load
    raise e
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 73, in load
    prog = converter.convert()
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/converter.py", line 227, in convert
    convert_nodes(self.context, self.graph)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 58, in convert_nodes
    _add_op(context, node)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 207, in constant
    const = _construct_constant(val, name)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/ops.py", line 196, in _construct_constant
    return mb.const(mode=mode, val=val, name=name)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/ops/registry.py", line 62, in add_op
    return cls._add_op(op_cls, **kwargs)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/builder.py", line 189, in _add_op
    new_op.type_value_inference()
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/operation.py", line 240, in type_value_inference
    output_types = self.type_inference()
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/ops/defs/control_flow.py", line 140, in type_inference
    builtin_type, _ = self._get_type_val(self.val.val)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/ops/defs/control_flow.py", line 180, in _get_type_val
    _, builtin_type = numpy_val_to_builtin_val(value)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/types/type_mapping.py", line 262, in numpy_val_to_builtin_val
    builtintype = numpy_type_to_builtin_type(npval.dtype)
  File "/home/liyang/.local/lib/python3.8/site-packages/coremltools/converters/mil/mil/types/type_mapping.py", line 232, in numpy_type_to_builtin_type
    raise TypeError("Unsupported numpy type: %s" % (nptype))
TypeError: Unsupported numpy type: float32

• workround

pip3 install numpy==1.19.5 pycocotools==2.0.0

pytorch版本问题

• torch == 1.9.0 && numpy == 1.19.5
• coremltools == 0.4.1
• error:

File "/opt/public/airlock/deepvac/deepvac/cast/coreml.py", line 60, in process
    coreml_model = coremltools.convert(model=model, inputs=[input], 
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/_converters_entry.py", line 175, in convert
    mlmodel = mil_convert(
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/converter.py", line 128, in mil_convert
    proto = mil_convert_to_proto(model, convert_from, convert_to,
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/converter.py", line 171, in mil_convert_to_proto
    prog = frontend_converter(model, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/converter.py", line 85, in __call__
    return load(*args, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/load.py", line 70, in load
    converter = TorchConverter(torchscript, inputs, outputs, cut_at_symbols)
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/converter.py", line 146, in __init__
    self.graph = InternalTorchIRGraph(
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/internal_graph.py", line 241, in __init__
    new_node = InternalTorchIRNode(raw_node, parent=self)
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/internal_graph.py", line 140, in __init__
    self.attr = {
  File "/opt/conda/lib/python3.8/site-packages/coremltools/converters/mil/frontend/torch/internal_graph.py", line 141, in <dictcomp>
    name: getattr(node, node.kindOf(name))(name)
AttributeError: 'torch._C.Node' object has no attribute 'ival'

• workround: pytorch降级

torch.jit.script导出模型转换coreml模型问题

• error
  ···
  temporary: the only valid use of a module is looking up an attribute but found = prim::SetAttr[name="num_batches_tracked"]
  ···
• workround

1. 采用torch.jit.trace导出模型进行coreml转换
2. 对conv和bn操作进行合并操作 fuse_conv_and_bn
3. 利用coreml API对不支持操作进行注册
4. 相关问题