def make_request(model_name,
inputs: Iterable[Union[np.ndarray, torch.Tensor]],
meta=None):
inputs = list(inputs)
example = inputs[0]
if meta is None:
meta = dict()
if isinstance(example, np.ndarray):
to_byte = bytes
torch_flag = False
elif isinstance(example, torch.Tensor):
to_byte = compose(bytes, torch.Tensor.numpy)
torch_flag = True
else:
raise ValueError(
'Argument `image` is expected to be an iterative numpy array, or an iterative torch Tensor'
)
raw_input = list(map(to_byte, inputs))
shape = example.shape
dtype = type_to_data_type(example.dtype).value
meta = json_update(
{
'shape': shape,
'dtype': dtype,
'torch_flag': torch_flag
}, meta)
return InferRequest(model_name=model_name,
raw_input=raw_input,
meta=json.dumps(meta))
def infer(self, input_batch):
example = input_batch[0]
meta = dict()
raw_input = list(map(bytes, input_batch))
shape = example.shape
dtype = type_to_data_type(example.dtype).value
meta = json_update({'shape': shape, 'dtype': dtype, 'torch_flag': True}, meta)
self.stub.Infer(InferRequest(model_name='resnet50', raw_input=raw_input, meta=json.dumps(meta)))