[模型转换] 待实现的代码

[nxdong]

运行命令：

model_transform.py --model_name glm_block_0 --model_def /workspace/model/glm_block_0.onnx --mlir glm_block_0.mlir

报错：

SOPHGO Toolchain v1.3.beta.1-65-g0d07945a-20230918
2023/09/18 18:36:07 - INFO : 
         _____________________________________________________ 
        | preprocess:                                           |
        |   (x - mean) * scale                                  |
        '-------------------------------------------------------'
  config Preprocess args : 
        resize_dims           : same to net input dims
        keep_aspect_ratio     : False
        keep_ratio_mode       : letterbox
        pad_value             : 0
        pad_type              : center
        --------------------------
        mean                  : [0.0, 0.0, 0.0]
        scale                 : [1.0, 1.0, 1.0]
        --------------------------
        pixel_format          : bgr
        channel_format        : nchw

Input_shape assigned
ConstantFolding finished
skip_fuse_bn: False
Onnxsim opt finished
ConstantFolding finished
num_const: 2
Traceback (most recent call last):
  File "/workspace/tpu-mlir/python/tools/model_transform.py", line 273, in <module>
    tool.model_transform(args.mlir, args.add_postprocess)
  File "/workspace/tpu-mlir/python/tools/model_transform.py", line 50, in model_transform
    self.converter.generate_mlir(mlir_origin)
  File "/workspace/tpu-mlir/python/transform/OnnxConverter.py", line 526, in generate_mlir
    self.onnxop_factory.get(n.op_type, lambda x: NoneAndRaise(x))(n)
  File "/workspace/tpu-mlir/python/transform/OnnxConverter.py", line 234, in <lambda>
    "Where": lambda node: self.convert_where_op(node),
  File "/workspace/tpu-mlir/python/transform/OnnxConverter.py", line 2009, in convert_where_op
    assert (0)  # TODO: to be implement
AssertionError

[nxdong]

这个问题可以通过在 tpu-mlir 项目的 ./python/transform/OnnxConverter.py 文件的 convert_where_op 函数内 增加

        elif num_const == 2:
            x_is_const = False
            y_is_const = False
            if self.isScalar(tbrn):
                x_is_const = True
                x_const_val = self.getScalar(tbrn)
                t_opd = self.mlir.none_op
            else:
                t_opd = tbrn_opd
                x_const_val = 0
            if self.isScalar(fbrn):
                y_is_const = True
                y_const_val = self.getScalar(fbrn)
                f_opd = self.mlir.none_op
            else:
                f_opd = fbrn_opd
                y_const_val = 0
            new_op = top.WhereOp(self.unranked_type,
                                 cond_opd,
                                 t_opd,
                                 f_opd,
                                 x_is_const=x_is_const,
                                 y_is_const=y_is_const,
                                 x_const_val=x_const_val,
                                 y_const_val=y_const_val,
                                 loc=self.get_loc("{}_{}".format(onnx_node.name,
                                                                 onnx_node.op_type)),
                                 ip=self.mlir.insert_point).output
        else:
            print("num_const:", num_const)
            assert (0)  # TODO: to be implement

来跳过。（是否合理待验证）