未验证 提交 2efdf0ef 编写于 作者: Y Yi Liu 提交者: GitHub

update en document of shard_index_op (#19963)

test=develop
test=document_fix
上级 766bd529
...@@ -296,7 +296,7 @@ paddle.fluid.layers.deformable_conv (ArgSpec(args=['input', 'offset', 'mask', 'n ...@@ -296,7 +296,7 @@ paddle.fluid.layers.deformable_conv (ArgSpec(args=['input', 'offset', 'mask', 'n
paddle.fluid.layers.unfold (ArgSpec(args=['x', 'kernel_sizes', 'strides', 'paddings', 'dilations', 'name'], varargs=None, keywords=None, defaults=(1, 0, 1, None)), ('document', '3f884662ad443d9ecc2b3734b4f61ad6')) paddle.fluid.layers.unfold (ArgSpec(args=['x', 'kernel_sizes', 'strides', 'paddings', 'dilations', 'name'], varargs=None, keywords=None, defaults=(1, 0, 1, None)), ('document', '3f884662ad443d9ecc2b3734b4f61ad6'))
paddle.fluid.layers.deformable_roi_pooling (ArgSpec(args=['input', 'rois', 'trans', 'no_trans', 'spatial_scale', 'group_size', 'pooled_height', 'pooled_width', 'part_size', 'sample_per_part', 'trans_std', 'position_sensitive', 'name'], varargs=None, keywords=None, defaults=(False, 1.0, [1, 1], 1, 1, None, 1, 0.1, False, None)), ('document', '99c03e3f249e36854f87dedaa17c8f35')) paddle.fluid.layers.deformable_roi_pooling (ArgSpec(args=['input', 'rois', 'trans', 'no_trans', 'spatial_scale', 'group_size', 'pooled_height', 'pooled_width', 'part_size', 'sample_per_part', 'trans_std', 'position_sensitive', 'name'], varargs=None, keywords=None, defaults=(False, 1.0, [1, 1], 1, 1, None, 1, 0.1, False, None)), ('document', '99c03e3f249e36854f87dedaa17c8f35'))
paddle.fluid.layers.filter_by_instag (ArgSpec(args=['ins', 'ins_tag', 'filter_tag', 'is_lod'], varargs=None, keywords=None, defaults=None), ('document', '7703a2088af8de4128b143ff1164ca4a')) paddle.fluid.layers.filter_by_instag (ArgSpec(args=['ins', 'ins_tag', 'filter_tag', 'is_lod'], varargs=None, keywords=None, defaults=None), ('document', '7703a2088af8de4128b143ff1164ca4a'))
paddle.fluid.layers.shard_index (ArgSpec(args=['input', 'index_num', 'nshards', 'shard_id', 'ignore_value'], varargs=None, keywords=None, defaults=(-1,)), ('document', '5786fdbba6753ecd6cbce5e6b0889924')) paddle.fluid.layers.shard_index (ArgSpec(args=['input', 'index_num', 'nshards', 'shard_id', 'ignore_value'], varargs=None, keywords=None, defaults=(-1,)), ('document', 'c4969dd6bf164f9e6a90414ea4f4e5ad'))
paddle.fluid.layers.hard_swish (ArgSpec(args=['x', 'threshold', 'scale', 'offset', 'name'], varargs=None, keywords=None, defaults=(6.0, 6.0, 3.0, None)), ('document', '6a5152a7015c62cb8278fc24cb456459')) paddle.fluid.layers.hard_swish (ArgSpec(args=['x', 'threshold', 'scale', 'offset', 'name'], varargs=None, keywords=None, defaults=(6.0, 6.0, 3.0, None)), ('document', '6a5152a7015c62cb8278fc24cb456459'))
paddle.fluid.layers.mse_loss (ArgSpec(args=['input', 'label'], varargs=None, keywords=None, defaults=None), ('document', 'd9ede6469288636e1b3233b461a165c9')) paddle.fluid.layers.mse_loss (ArgSpec(args=['input', 'label'], varargs=None, keywords=None, defaults=None), ('document', 'd9ede6469288636e1b3233b461a165c9'))
paddle.fluid.layers.data (ArgSpec(args=['name', 'shape', 'append_batch_size', 'dtype', 'lod_level', 'type', 'stop_gradient'], varargs=None, keywords=None, defaults=(True, 'float32', 0, VarType.LOD_TENSOR, True)), ('document', '9d7806e31bdf727c1a23b8782a09b545')) paddle.fluid.layers.data (ArgSpec(args=['name', 'shape', 'append_batch_size', 'dtype', 'lod_level', 'type', 'stop_gradient'], varargs=None, keywords=None, defaults=(True, 'float32', 0, VarType.LOD_TENSOR, True)), ('document', '9d7806e31bdf727c1a23b8782a09b545'))
......
...@@ -14122,52 +14122,46 @@ def deformable_roi_pooling(input, ...@@ -14122,52 +14122,46 @@ def deformable_roi_pooling(input,
def shard_index(input, index_num, nshards, shard_id, ignore_value=-1): def shard_index(input, index_num, nshards, shard_id, ignore_value=-1):
""" """
This layer creates the sharded index for input. This layers is used in This function recomputes the `input` indices according to the offset of the
model- and data- parallel mixed training generally, in which the index shard. The length of the indices is evenly divided into N shards, and if
data (usually the label) should be recaculated in each trainer according the `shard_id` matches the shard with the input index inside, the index is
to recomputed on the basis of the shard offset, elsewise it is set to
`ignore_value`. The detail is as follows:
.. math:: ::
assert index_num % nshards == 0 shard_size = (index_num + nshards - 1) // nshards
y = x % shard_size if x // shard_size == shard_id else ignore_value
shard_size = index_num / nshards
y = x % shard_size if x / shard_size == shard_id else ignore_value
We take the distributed one-hot representation to show what this layer is NOTE: If the length of indices cannot be evely divided by the shard number,
used for. The distributed one-hot representation is seperated into multiple the size of the last shard will be less than the calculated `shard_size`
shards, and each shard is filling zeros except the one with the index
inside. In order to create these sharded representation in each trainer,
the original index should be recalculated (i.e. sharded) before.
Examples: Examples:
::
X is a Tensor of integer values: Input:
X.shape = [4, 1] X.shape = [4, 1]
X.data = [[1], [6], [12], [19]] X.data = [[1], [6], [12], [19]]
index_num = 20
nshards = 2
ignore_value = -1
suppose index_num = 20 and nshards = 2, then we get shard_size = 10 if shard_id == 0, we get:
if shard_id == 0, we get the Out:
Out.shape = [4, 1] Out.shape = [4, 1]
Out.data = [[1], [6], [-1], [-1]] Out.data = [[1], [6], [-1], [-1]]
if shard_id == 1, we get the Out: if shard_id == 1, we get:
Out.shape = [4, 1] Out.shape = [4, 1]
Out.data = [[-1], [-1], [2], [9]] Out.data = [[-1], [-1], [2], [9]]
the default `ignore_value` -1 is used in this example.
Args: Args:
input(Variable): Input indices, last dimension must be 1. - **input** (Variable): Input indices, last dimension must be 1.
index_num(scalar): An interger defining the range of the index. - **index_num** (scalar): An interger defining the range of the index.
nshards(scalar): The number of shards - **nshards** (scalar): The number of shards
shard_id(scalar): The index of the current shard - **shard_id** (scalar): The index of the current shard
ignore_value(scalar): An ingeter value out of sharded index range - **ignore_value** (scalar): An ingeter value out of sharded index range
Returns: Returns:
Variable: The shard index of input. Variable: The sharded index of input.
Examples: Examples:
.. code-block:: python .. code-block:: python
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册