SwinTransformer模型转换失败

[UnlimitedWand]

问题描述

SwinTransformer模型转换失败，使用本仓库转换代码以及模型

运行环境，aarch64， torch是cpu版本，2.5.1和1.10.2都失败，同样的报错
PaddlePaddle = 2.6.1:
X2Paddle = 1.5.0

• 错误信息
• warnings.warn(
  Fail to generate inference model! Problem happened while export inference model from python code '/work/code/X2Paddle/test_benchmark/PyTorch/SwinTransformer/pd_model_trace/x2paddle_code.py';

===================Error Information===============
Traceback (most recent call last):
File "/work/code/X2Paddle/test_benchmark/PyTorch/SwinTransformer/convert_trace.py", line 27, in
pytorch2paddle(torch_module,
File "/opt/py310/lib/python3.10/site-packages/x2paddle/convert.py", line 398, in pytorch2paddle
mapper.paddle_graph.gen_model(save_dir,
File "/opt/py310/lib/python3.10/site-packages/x2paddle/core/program.py", line 308, in gen_model
raise e
File "/opt/py310/lib/python3.10/site-packages/x2paddle/core/program.py", line 302, in gen_model
self.dygraph2static(save_dir, input_shapes, input_types)
File "/opt/py310/lib/python3.10/site-packages/x2paddle/core/program.py", line 584, in dygraph2static
paddle.jit.save(static_model,
File "/opt/py310/lib/python3.10/site-packages/decorator.py", line 232, in fun
return caller(func, *(extras + args), **kw)
File "/opt/py310/lib/python3.10/site-packages/paddle/base/wrapped_decorator.py", line 26, in impl
return wrapped_func(*args, **kwargs)
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/api.py", line 809, in wrapper
func(layer, path, input_spec, **configs)
File "/opt/py310/lib/python3.10/site-packages/decorator.py", line 232, in fun
return caller(func, *(extras + args), **kw)
File "/opt/py310/lib/python3.10/site-packages/paddle/base/wrapped_decorator.py", line 26, in impl
return wrapped_func(*args, **kwargs)
File "/opt/py310/lib/python3.10/site-packages/paddle/base/dygraph/base.py", line 68, in impl
return func(*args, **kwargs)
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/api.py", line 1104, in save
static_func.concrete_program_specify_input_spec(
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/dy2static/program_translator.py", line 986, in concrete_program_specify_input_spec
concrete_program, _ = self.get_concrete_program(
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/dy2static/program_translator.py", line 875, in get_concrete_program
concrete_program, partial_program_layer = self._program_cache[
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/dy2static/program_translator.py", line 1648, in getitem
self._caches[item_id] = self._build_once(item)
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/dy2static/program_translator.py", line 1575, in _build_once
concrete_program = ConcreteProgram.from_func_spec(
File "/opt/py310/lib/python3.10/site-packages/decorator.py", line 232, in fun
return caller(func, *(extras + args), **kw)
File "/opt/py310/lib/python3.10/site-packages/paddle/base/wrapped_decorator.py", line 26, in impl
return wrapped_func(*args, **kwargs)
File "/opt/py310/lib/python3.10/site-packages/paddle/base/dygraph/base.py", line 68, in impl
return func(*args, **kwargs)
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/dy2static/program_translator.py", line 1346, in from_func_spec
error_data.raise_new_exception()
File "/opt/py310/lib/python3.10/site-packages/paddle/jit/dy2static/error.py", line 452, in raise_new_exception
raise new_exception from None
AssertionError: In transformed code:

File "/work/code/X2Paddle/test_benchmark/PyTorch/SwinTransformer/pd_model_trace/x2paddle_code.py", line 545, in forward
    layers_0_downsample_reduction_weight = self.layers_0_downsample_reduction_weight
    x274 = paddle.matmul(x=x272, y=layers_0_downsample_reduction_weight, transpose_y=True)
    x274 = paddle.add(x=x274, y=x47)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
    x276 = 0
    x277 = 2

File "/opt/py310/lib/python3.10/site-packages/paddle/tensor/math.py", line 699, in add
    return _elementwise_op(LayerHelper('elementwise_add', **locals()))
File "/opt/py310/lib/python3.10/site-packages/paddle/tensor/math.py", line 580, in _elementwise_op
    assert y is not None, f'y cannot be None in {original_op_type}'

AssertionError: y cannot be None in elementwise_add

• 模型来源
  本仓库提供的链接

[megemini]

@UnlimitedWand 是测试的 https://github.com/PaddlePaddle/X2Paddle/tree/develop/test_benchmark/PyTorch/SwinTransformer 这个吗？

最近在适配 Paddle 3.0，可以拉一下最新的 dev 版本的 x2paddle，用 Paddle 3.0 再试一下 ～

或者参考一下这个 PR #1079