OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2025-12-07 00:20:20 +01:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

Update ops-related pbtxt files.

Browse Source 
PiperOrigin-RevId: 158325293
This commit is contained in:
A. Unique TensorFlower 2017-06-07 15:17:52 -07:00 committed by TensorFlower Gardener
parent 0a9d2dac08
commit 4e529f0f16
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/core/ops/ops.pbtxt
View File

@ -26621,7 +26621,7 @@ op {
description: "The gradient source string, used to decide which gradient TensorArray\nto return."
}
summary: "Creates a TensorArray for storing the gradients of values in the given handle."
description: "If the given TensorArray gradient already exists, returns a reference to it.\n\nLocks the size of the original TensorArray by disabling its dynamic size flag.\n\n**A note about the input flow_in:**\n\nThe handle flow_in forces the execution of the gradient lookup to occur\nonly after certain other operations have occurred. For example, when\nthe forward TensorArray is dynamically sized, writes to this TensorArray\nmay resize the object. The gradient TensorArray is statically sized based\non the size of the forward TensorArray when this operation executes.\nFurthermore, the size of the forward TensorArray is frozen by this call.\nAs a result, the flow is used to ensure that the call to generate the gradient\nTensorArray only happens after all writes are executed.\n\nIn the case of dynamically sized TensorArrays, gradient computation should\nonly be performed on read operations that have themselves been chained via\nflow to occur only after all writes have executed. That way the final size\nof the forward TensorArray is known when this operation is called.\n\n**A note about the source attribute:**\n\nTensorArray gradient calls use an accumulator TensorArray object. If\nmultiple gradients are calculated and run in the same session, the multiple\ngradient nodes may accidentally flow throuth the same accumulator TensorArray.\nThis double counts and generally breaks the TensorArray gradient flow.\n\nThe solution is to identify which gradient call this particular\nTensorArray gradient is being called in. This is performed by identifying\na unique string (e.g. \"gradients\", \"gradients_1\", ...) from the input\ngradient Tensor\'s name. This string is used as a suffix when creating\nthe TensorArray gradient object here (the attribute `source`).\n\nThe attribute `source` is added as a suffix to the forward TensorArray\'s\nname when performing the creation / lookup, so that each separate gradient\ncalculation gets its own TensorArray accumulator."
description: "If the given TensorArray gradient already exists, returns a reference to it.\n\nLocks the size of the original TensorArray by disabling its dynamic size flag.\n\n**A note about the input flow_in:**\n\nThe handle flow_in forces the execution of the gradient lookup to occur\nonly after certain other operations have occurred. For example, when\nthe forward TensorArray is dynamically sized, writes to this TensorArray\nmay resize the object. The gradient TensorArray is statically sized based\non the size of the forward TensorArray when this operation executes.\nFurthermore, the size of the forward TensorArray is frozen by this call.\nAs a result, the flow is used to ensure that the call to generate the gradient\nTensorArray only happens after all writes are executed.\n\nIn the case of dynamically sized TensorArrays, gradient computation should\nonly be performed on read operations that have themselves been chained via\nflow to occur only after all writes have executed. That way the final size\nof the forward TensorArray is known when this operation is called.\n\n**A note about the source attribute:**\n\nTensorArray gradient calls use an accumulator TensorArray object. If\nmultiple gradients are calculated and run in the same session, the multiple\ngradient nodes may accidentally flow through the same accumulator TensorArray.\nThis double counts and generally breaks the TensorArray gradient flow.\n\nThe solution is to identify which gradient call this particular\nTensorArray gradient is being called in. This is performed by identifying\na unique string (e.g. \"gradients\", \"gradients_1\", ...) from the input\ngradient Tensor\'s name. This string is used as a suffix when creating\nthe TensorArray gradient object here (the attribute `source`).\n\nThe attribute `source` is added as a suffix to the forward TensorArray\'s\nname when performing the creation / lookup, so that each separate gradient\ncalculation gets its own TensorArray accumulator."
is_stateful: true
}
op {