Symbolic shape inference support for pd_op.bce_loss, pd_op.sigmoid_cross_entropy_with_logits

[WintersMontagne10335]

PR Category

CINN

PR Types

Others

Description

Symbolic shape inference support for pd_op.bce_loss, pd_op.sigmoid_cross_entropy_with_logits

[paddle-bot[bot]]

你的PR提交成功，感谢你对开源项目的贡献! 请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。 Your PR has been submitted. Thanks for your contribution! Please wait for the result of CI firstly. See Paddle CI Manual for details.

[WintersMontagne10335]

pd_op.bce_loss, pd_op.distribute_fpn_proposals, pd_op.sigmoid_cross_entropy_with_logits 对应的单测存在报错如下：

ValueError: (InvalidArgument) Each value of attribute 'shape' is expected to be no less than 0. But received: shape[0] = -1; shape = [-1, 256, 7, 7].
  [Hint: Expected shape[i] >= 0, but received shape[i]:-1 < 0:0.] (at /home/aistudio/Paddle/paddle/fluid/operators/fill_constant_op.cc:41)
  [operator < fill_constant > error]

所以无法验证三者的 shape 推导是否正确。目前正在尝试解决 fill_constant 的问题，进展如下：

{Out=['index_select_0.tmp_0']} = gather(inputs={Axis=[], Index=['range_0.tmp_0'], X=['one_hot_v2_0.tmp_0']}, axis = 2, op_device = , op_namescope = /, op_role = 0, op_role_var = [], with_quant_attr = False)

{Out=['tmp_10']} = elementwise_mul(inputs={X=['index_select_0.tmp_0'], Y=['unsqueeze2_0.tmp_0']}, axis = -1, op_device = , op_namescope = /, op_role = 0, op_role_var = [], with_quant_attr = False)

 'one_hot_v2_0.tmp_0': var one_hot_v2_0.tmp_0 : LOD_TENSOR.shape(1, 3549, 81).dtype(float32).stop_gradient(True)

# 而 index_select_0.tmp_0 的shape含有 -1

初步推断是 gather 的 shape 推导有问题，但是在 GatherOpInferSymbolicShape 添加 VLOG ，日志中却并没有输出相关信息。

正在查看其它与 shape 推导有关的函数。

[paddle-ci-bot[bot]]

Sorry to inform you that 2687559's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.

[WintersMontagne10335]

pd_op.bce_loss, pd_op.distribute_fpn_proposals, pd_op.sigmoid_cross_entropy_with_logits 对应的单测存在报错如下：

ValueError: (InvalidArgument) Each value of attribute 'shape' is expected to be no less than 0. But received: shape[0] = -1; shape = [-1, 256, 7, 7].
  [Hint: Expected shape[i] >= 0, but received shape[i]:-1 < 0:0.] (at /home/aistudio/Paddle/paddle/fluid/operators/fill_constant_op.cc:41)
  [operator < fill_constant > error]

所以无法验证三者的 shape 推导是否正确。目前正在尝试解决 fill_constant 的问题，进展如下：

{Out=['index_select_0.tmp_0']} = gather(inputs={Axis=[], Index=['range_0.tmp_0'], X=['one_hot_v2_0.tmp_0']}, axis = 2, op_device = , op_namescope = /, op_role = 0, op_role_var = [], with_quant_attr = False)

{Out=['tmp_10']} = elementwise_mul(inputs={X=['index_select_0.tmp_0'], Y=['unsqueeze2_0.tmp_0']}, axis = -1, op_device = , op_namescope = /, op_role = 0, op_role_var = [], with_quant_attr = False)

 'one_hot_v2_0.tmp_0': var one_hot_v2_0.tmp_0 : LOD_TENSOR.shape(1, 3549, 81).dtype(float32).stop_gradient(True)

# 而 index_select_0.tmp_0 的shape含有 -1

初步推断是 gather 的 shape 推导有问题，但是在 GatherOpInferSymbolicShape 添加 VLOG ，日志中却并没有输出相关信息。

正在查看其它与 shape 推导有关的函数。

是低级错误，尴尬

[paddle-ci-bot[bot]]

Sorry to inform you that 534197d's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.

[paddle-ci-bot[bot]]

Sorry to inform you that 62c9b16's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.