mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
Revert "Fix typos."
This reverts commit 4622b3395276b37e10141fab43ffea33941ca0c2.
This commit is contained in:
Gregory Chanan
gchanan
@ -542,7 +542,7 @@ mark_as_advanced(
|
||||
# Makefile and similar generators don't define CMAKE_CONFIGURATION_TYPES, so we
|
||||
# need to add another entry for the CMAKE_BUILD_TYPE. We also need to add the
|
||||
# standerd set of 4 build types (Debug, MinSizeRel, Release, and RelWithDebInfo)
|
||||
# for completeness. We need run this loop in order to accommodate the addition
|
||||
# for completeness. We need run this loop in order to accomodate the addition
|
||||
# of extra configuration types. Duplicate entries will be removed by
|
||||
# REMOVE_DUPLICATES.
|
||||
set(CUDA_configuration_types ${CMAKE_CONFIGURATION_TYPES} ${CMAKE_BUILD_TYPE} Debug MinSizeRel Release RelWithDebInfo)
|
||||
|
||||
@ -168,7 +168,7 @@ if(CUDA_VERSION VERSION_LESS "3.0")
|
||||
# CMake policy 0007 NEW states that empty list elements are not
|
||||
# ignored. I'm just setting it to avoid the warning that's printed.
|
||||
cmake_policy(SET CMP0007 NEW)
|
||||
# Note that this will remove all occurrences of -G.
|
||||
# Note that this will remove all occurances of -G.
|
||||
list(REMOVE_ITEM depends_CUDA_NVCC_FLAGS "-G")
|
||||
cmake_policy(POP)
|
||||
endif()
|
||||
|
||||
@ -2545,7 +2545,7 @@ class TestNN(NNTestCase):
|
||||
self.assertEqual(input_cpu.grad, input_cuda.grad)
|
||||
self.assertEqual(grid_cpu.grad, grid_cuda.grad)
|
||||
|
||||
# check that zero-dimensional input strides don't error out
|
||||
# check that zero-dimensional input strides dont error out
|
||||
base_input = torch.randn(C, IH, IW)
|
||||
input_cpu = Variable(base_input.expand(input_cuda.size()), requires_grad=True)
|
||||
grid_cpu = Variable(torch.randn(N, H, W, 2), requires_grad=True)
|
||||
|
||||
@ -4009,7 +4009,7 @@ class TestTorch(TestCase):
|
||||
expected = torch.arange(1, 126).view(5, 5, 5)[:, 1]
|
||||
self.assertEqual(torch.from_numpy(x), expected)
|
||||
|
||||
# check zero dimensional
|
||||
# check zero dimentional
|
||||
x = np.zeros((0, 2))
|
||||
self.assertRaises(RuntimeError, lambda: torch.from_numpy(x))
|
||||
|
||||
|
||||
@ -26,7 +26,7 @@ class Backend(object):
|
||||
self.loading_lock = threading.Lock()
|
||||
|
||||
def load(self):
|
||||
# This looks a little weird, but it's necessary for thread safe loading.
|
||||
# This looks a little weird, but it's neccesary for thread safe loading.
|
||||
# Loading the backend can take some time, so multiple threads can enter
|
||||
# the if clause. We have to ensure that only the first one to acquire
|
||||
# the lock will actually load the backend, and that the rest won't
|
||||
|
||||
@ -1817,7 +1817,7 @@ along a given dimension.
|
||||
If :attr:`dim` is not given, the last dimension of the `input` is chosen.
|
||||
|
||||
A tuple of `(values, indices)` is returned, where the `indices` is the indices
|
||||
of the kth-smallest element in the original `input` Tensor in dimension `dim`.
|
||||
of the kth-smallest element in the original `input` Tensor in dimention `dim`.
|
||||
|
||||
If :attr:`keepdim` is true, both the :attr:`values` and :attr:`indices` Tensors
|
||||
are the same size as :attr:`input`, except in the dimension :attr:`dim` where
|
||||
|
||||
@ -48,7 +48,7 @@ def backward(variables, grad_variables=None, retain_graph=None, create_graph=Non
|
||||
|
||||
The graph is differentiated using the chain rule. If any of ``variables``
|
||||
are non-scalar (i.e. their data has more than one element) and require
|
||||
gradient, the function additionally requires specifying ``grad_variables``.
|
||||
gradient, the function additionaly requires specifying ``grad_variables``.
|
||||
It should be a sequence of matching length, that contains gradient of
|
||||
the differentiated function w.r.t. corresponding variables (``None`` is an
|
||||
acceptable value for all variables that don't need gradient tensors).
|
||||
|
||||
@ -142,7 +142,7 @@ class _SelectionFunction(Function):
|
||||
has_all_reduce = True
|
||||
# additional_args is prepended before dim when calling the tensor
|
||||
# function. It's a no-op for subclasses other than kthvalue.
|
||||
# kthvalue not only requires us to pass a dim, but also precede it with k.
|
||||
# kthvalue not only requires us to pass a dim, but also preceed it with k.
|
||||
|
||||
@classmethod
|
||||
def forward(cls, ctx, input, dim=None, keepdim=None, additional_args=tuple()):
|
||||
|
||||
@ -130,7 +130,7 @@ class Variable(_C._VariableBase):
|
||||
|
||||
The graph is differentiated using the chain rule. If the variable is
|
||||
non-scalar (i.e. its data has more than one element) and requires
|
||||
gradient, the function additionally requires specifying ``gradient``.
|
||||
gradient, the function additionaly requires specifying ``gradient``.
|
||||
It should be a tensor of matching type and location, that contains
|
||||
the gradient of the differentiated function w.r.t. ``self``.
|
||||
|
||||
|
||||
@ -1,7 +1,7 @@
|
||||
#include "torch/csrc/cuda/THCP.h"
|
||||
|
||||
// Declare/Define the expansion functions that have THCState. Note that we
|
||||
// still need to define the CPU-type versions because the copy functions that
|
||||
// still need to define the CPU-type verisons because the copy functions that
|
||||
// copy from GPU to CPU type have a THCState.
|
||||
|
||||
#define CUDA_EXPAND 1
|
||||
|
||||
@ -1384,7 +1384,7 @@ static PyObject * THPTensor_(getValue)(THPTensor *self, PyObject *index)
|
||||
return THPUtils_(newReal)(THStorage_(get)(LIBRARY_STATE sresult, storage_offset));
|
||||
}
|
||||
}
|
||||
THPUtils_setError("An unknown error has occurred when indexing a tensor "
|
||||
THPUtils_setError("An unknown error has occured when indexing a tensor "
|
||||
"in THPTensor_(getValue). Please report this in a github issue at: "
|
||||
"https://github.com/pytorch/pytorch");
|
||||
return NULL;
|
||||
@ -1482,7 +1482,7 @@ static int THPTensor_(setValue)(THPTensor *self, PyObject *index, PyObject *valu
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
THPUtils_setError("An unknown error has occurred when indexing a tensor "
|
||||
THPUtils_setError("An unknown error has occured when indexing a tensor "
|
||||
"in THPTensor_(setValue). Please report this in a github issue at: "
|
||||
"https://github.com/pytorch/pytorch");
|
||||
return -1;
|
||||
|
||||
@ -98,7 +98,7 @@ PyObject * THPTensor_(toNumpy)(THPTensor *self, PyObject *args) {
|
||||
strides.get(), self->cdata->storage->data + self->cdata->storageOffset,
|
||||
0, NPY_ARRAY_ALIGNED | NPY_ARRAY_WRITEABLE | NPY_ARRAY_C_CONTIGUOUS, nullptr));
|
||||
if (!array) {
|
||||
THPUtils_setError("an error occurred during conversion to numpy array");
|
||||
THPUtils_setError("an error occured during conversion to numpy array");
|
||||
return NULL;
|
||||
}
|
||||
|
||||
@ -108,7 +108,7 @@ PyObject * THPTensor_(toNumpy)(THPTensor *self, PyObject *args) {
|
||||
Py_INCREF(self);
|
||||
if (PyArray_SetBaseObject((PyArrayObject*)(array.get()), (PyObject*)self) == -1) {
|
||||
Py_DECREF(self);
|
||||
THPUtils_setError("an error occurred during conversion to numpy array");
|
||||
THPUtils_setError("an error occured during conversion to numpy array");
|
||||
return NULL;
|
||||
}
|
||||
|
||||
|
||||
@ -138,7 +138,7 @@ def scatter(tensor, devices, chunk_sizes=None, dim=0, streams=None):
|
||||
dim (int, optional): A dimension along which to chunk the tensor.
|
||||
|
||||
Returns:
|
||||
A tuple containing chunks of the ``tensor``, spread across given
|
||||
A tuple containing chunks of the ``tensor``, spread accross given
|
||||
``devices``.
|
||||
"""
|
||||
if chunk_sizes is None:
|
||||
|
||||
@ -1,6 +1,6 @@
|
||||
"""
|
||||
torch.distributed provides an MPI-like interface for exchanging tensor
|
||||
data across multi-machine networks. It supports a few different backends
|
||||
data accross multi-machine networks. It supports a few different backends
|
||||
and initialization methods.
|
||||
"""
|
||||
import torch
|
||||
@ -105,7 +105,7 @@ class _DistributedRequest(object):
|
||||
def get_rank():
|
||||
"""Returns the rank of current process.
|
||||
|
||||
Rank is a unique identifier assigned to each process within a distributed
|
||||
Rank is a unique identifier assigned to each process withing a distributed
|
||||
group. They are always consecutive integers ranging from 0 to ``world_size``.
|
||||
"""
|
||||
assert torch.distributed._initialized
|
||||
@ -192,7 +192,7 @@ def broadcast(tensor, src, group=group.WORLD):
|
||||
|
||||
|
||||
def all_reduce(tensor, op=reduce_op.SUM, group=group.WORLD):
|
||||
"""Reduces the tensor data across all machines in such a way that all get
|
||||
"""Reduces the tensor data accross all machines in such a way that all get
|
||||
the final result.
|
||||
|
||||
After the call ``tensor`` is going to be bitwise identical in all processes.
|
||||
@ -210,7 +210,7 @@ def all_reduce(tensor, op=reduce_op.SUM, group=group.WORLD):
|
||||
|
||||
|
||||
def reduce(tensor, dst, op=reduce_op.SUM, group=group.WORLD):
|
||||
"""Reduces the tensor data across all machines.
|
||||
"""Reduces the tensor data accross all machines.
|
||||
|
||||
Only the process with rank ``dst`` is going to receive the final result.
|
||||
|
||||
|
||||
@ -111,7 +111,7 @@ class LookupTable(Module):
|
||||
if self.maxNorm is None:
|
||||
return
|
||||
|
||||
# copy input into _input, so _input is continuous.
|
||||
# copy input into _input, so _input is continous.
|
||||
# The copied _input will be modified in the C code.
|
||||
self._input.resize_(input.size()).copy_(input)
|
||||
row_idx = self._input
|
||||
|
||||
@ -21,7 +21,7 @@ class SpatialDropout(Module):
|
||||
|
||||
self.noise.bernoulli_(1 - self.p)
|
||||
# We expand the random dropouts to the entire feature map because the
|
||||
# features are likely correlated across the map and so the dropout
|
||||
# features are likely correlated accross the map and so the dropout
|
||||
# should also be correlated.
|
||||
self.output.mul_(self.noise.expand_as(input))
|
||||
else:
|
||||
|
||||
@ -30,8 +30,8 @@ class Sum(Module):
|
||||
|
||||
def updateGradInput(self, input, gradOutput):
|
||||
dimension = self._getPositiveDimension(input)
|
||||
# zero-strides don't work with MKL/BLAS, so
|
||||
# don't set self.gradInput to zero-stride tensor.
|
||||
# zero-strides dont work with MKL/BLAS, so
|
||||
# dont set self.gradInput to zero-stride tensor.
|
||||
# Instead, do a deepcopy.
|
||||
size = list(input.size())
|
||||
size[dimension] = 1
|
||||
|
||||
@ -19,7 +19,7 @@ class VolumetricDropout(Module):
|
||||
|
||||
self.noise.bernoulli_(1 - self.p)
|
||||
# We expand the random dropouts to the entire feature map because the
|
||||
# features are likely correlated across the map and so the dropout
|
||||
# features are likely correlated accross the map and so the dropout
|
||||
# should also be correlated.
|
||||
self.output.mul_(self.noise.expand_as(input))
|
||||
else:
|
||||
|
||||
@ -12,7 +12,7 @@ def asgd(opfunc, x, config, state=None):
|
||||
eta_t = eta0 / (1 + lambda eta0 t) ^ 0.75
|
||||
mu_t = 1/max(1,t-t0)
|
||||
|
||||
implements ASGD algorithm as in L.Bottou's sgd-2.0
|
||||
implements ASGD algoritm as in L.Bottou's sgd-2.0
|
||||
|
||||
ARGS:
|
||||
|
||||
|
||||
@ -38,7 +38,7 @@ public:
|
||||
|
||||
#undef DEFINE_IMPLICIT_CTOR
|
||||
|
||||
// return a new scalar that is guaranteed to be not backed by a tensor.
|
||||
// return a new scalar that is guarenteed to be not backed by a tensor.
|
||||
Scalar local() {
|
||||
if (Tag::HAS_t != tag) {
|
||||
return *this;
|
||||
|
||||
@ -154,7 +154,7 @@ IF(C_SSE3_FOUND)
|
||||
MESSAGE(STATUS "SSE3 Found")
|
||||
SET(CMAKE_C_FLAGS "${C_SSE3_FLAGS} -DUSE_SSE3 ${CMAKE_C_FLAGS}")
|
||||
ENDIF(C_SSE3_FOUND)
|
||||
# we don't set -mavx and -mavx2 flags globally, but only for specific files
|
||||
# we dont set -mavx and -mavx2 flags globally, but only for specific files
|
||||
# however, we want to enable the AVX codepaths, so we still need to
|
||||
# add USE_AVX and USE_AVX2 macro defines
|
||||
IF(C_AVX_FOUND)
|
||||
|
||||
@ -1345,7 +1345,7 @@ static void THFloatVector_divs_VSX(float *y, const float*x, const float c, const
|
||||
// TODO
|
||||
//
|
||||
//
|
||||
// Finished running all tests. All tests PASSED.
|
||||
// Finished runnning all tests. All tests PASSED.
|
||||
//
|
||||
//------------------------------------------------
|
||||
#ifdef RUN_VSX_TESTS
|
||||
@ -2509,7 +2509,7 @@ int main()
|
||||
|
||||
|
||||
|
||||
printf("Finished running all tests. All tests PASSED.\n");
|
||||
printf("Finished runnning all tests. All tests PASSED.\n");
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
@ -22,7 +22,7 @@
|
||||
THC_API THAllocator THCCachingHostAllocator;
|
||||
|
||||
// Records an event in the specified stream. The allocation 'ptr' will not be
|
||||
// re-used until the event has occurred.
|
||||
// re-used until the event has occured.
|
||||
THC_API cudaError_t THCCachingHostAllocator_recordEvent(void *ptr, THCStream *stream);
|
||||
|
||||
// Releases cached pinned memory allocations via cudaHostFree
|
||||
|
||||
@ -663,7 +663,7 @@ THC_API void THCudaTensor_conv2DRevgerm(THCState *state, THCudaTensor *output, f
|
||||
float *output_data = THCudaTensor_data(state, output);
|
||||
|
||||
// kernel is called multiple times
|
||||
// (the arbitrary split below is just here to make sure we don't go over 256 threads)
|
||||
// (the arbitrary split below is just here to make sure we dont go over 256 threads)
|
||||
for (int sl=0; sl<nbatch; sl+=6) {
|
||||
// auto compute nb of blocks and threads
|
||||
dim3 blocks(nKernelPlane, nInputPlane);
|
||||
|
||||
@ -245,7 +245,7 @@ __global__ void computeMode(
|
||||
// Finally, we need to find the "an" index of the mode in the input Tensor. The API does
|
||||
// not constrain which index we pick, so it can be any of the indices that contain the mode.
|
||||
// We will do a reduction to find the index. We go back to using the (index, flag) buffer
|
||||
// arrangement. First, we mark indices that are equal to the mode, i.e B[i] = true if
|
||||
// arrangment. First, we mark indices that are equal to the mode, i.e B[i] = true if
|
||||
// input[i] == mode, and initialize C[i] to be the index
|
||||
//
|
||||
// Again we reduce 2 elements in the thread's registers prior to the block-wide reduction
|
||||
|
||||
@ -105,7 +105,7 @@ __global__ void cunn_LookupTable_accGradParametersKernel(
|
||||
int idx = blockIdx.x * 4 + threadIdx.y;
|
||||
|
||||
// Each warp is responsible for an input into the LookupTable.
|
||||
// If the preceding input has the same as this input, then the warp
|
||||
// If the preceeding input has the same as this input, then the warp
|
||||
// exits immediately. The warp also processes subsequent inputs with the
|
||||
// same value.
|
||||
//
|
||||
|
||||
@ -156,7 +156,7 @@ void THNN_(SparseLinear_accGradParameters)(
|
||||
THCTensor_(select)(state, sel, input, 1, 0); // rowInds
|
||||
THCTensor_(select)(state, cols, input, 1, 1); // colInds
|
||||
THCTensor_(cadd)(state, buf, sel, batchnum, cols); // colInds * buatchdim + rowInds
|
||||
THCTensor_(sort)(state, buf, inds, buf, 0, 0); // Indices are now in ind
|
||||
THCTensor_(sort)(state, buf, inds, buf, 0, 0); // Indicies are now in ind
|
||||
THCTensor_(indexSelect)(state, buf, input, 0, inds);
|
||||
|
||||
THCTensor_(resize1d)(state, values, nnz);
|
||||
|
||||
@ -70,7 +70,7 @@ void Store::StoreDeamon::deamon() {
|
||||
try {
|
||||
query(rank);
|
||||
} catch (...) {
|
||||
// There was an error when processing query. Probably an exception occurred in
|
||||
// There was an error when processing query. Probably an exception occured in
|
||||
// recv/send what would indicate that socket on the other side has been closed.
|
||||
// If the closing was due to normal exit, then the store should exit too.
|
||||
// Otherwise, if it was different exception, other processes will get
|
||||
@ -190,7 +190,7 @@ void Store::wait(const std::vector<std::string>& keys) {
|
||||
for (std::size_t i = 0; i < nkeys; i++) {
|
||||
send_string(_socket, keys[i], (i != (nkeys - 1)));
|
||||
}
|
||||
// after sending the query, wait for a 'stop_waiting' response
|
||||
// after sending the query, wait for a 'stop_waiting' reponse
|
||||
QueryType qr;
|
||||
recv_bytes<QueryType>(_socket, &qr, 1);
|
||||
if (qr != QueryType::STOP_WAITING)
|
||||
|
||||
@ -177,7 +177,7 @@ std::vector<MulticastMessage> getMessages(struct sockaddr* addr, rank_type world
|
||||
|
||||
if (recv_message_str == packed_msg) continue; // ignore multicast loopback
|
||||
|
||||
// We should ignore messages coming from different group
|
||||
// We should ignore messages comming from different group
|
||||
auto recv_msg = MulticastMessage(recv_message_str);
|
||||
if (recv_msg.group_name != group_name) {
|
||||
continue;
|
||||
|
||||
@ -23,7 +23,7 @@ overview of these domains and how they interact.
|
||||
|
||||
## Overview
|
||||
|
||||
Gloo algorithms are collective algorithms, meaning they can run in
|
||||
Gloo algorithms are collective algoritms, meaning they can run in
|
||||
parallel across two or more processes/machines. To be able to execute
|
||||
across multiple machines, they first need to find each other. We call
|
||||
this _rendezvous_ and it is the first thing to address when
|
||||
|
||||
@ -58,7 +58,7 @@ rendezvous to a particular namespace.
|
||||
|
||||
There are many more key/value stores that can be used for rendezvous
|
||||
(e.g. [etcd](https://coreos.com/etcd) or [ZooKeeper](https://zookeeper.apache.org/)).
|
||||
As long as a C or C++ interface for your store of choice is available,
|
||||
As long as a C or C++ interface for your store of choice is availabe,
|
||||
is relatively easy to hook it up to the Gloo rendezvous process.
|
||||
See the `gloo::rendezvous::Store` abstract base class for the interface to implement.
|
||||
|
||||
|
||||
@ -58,7 +58,7 @@ protected:
|
||||
bytes_received += step_received;
|
||||
buffer += step_received;
|
||||
} else if (pfd.revents & (POLLERR | POLLHUP)) {
|
||||
throw std::runtime_error("An error occurred while waiting for the data");
|
||||
throw std::runtime_error("An error occured while waiting for the data");
|
||||
} else {
|
||||
throw std::runtime_error("Shared memory manager connection has timed out");
|
||||
}
|
||||
|
||||
@ -46,7 +46,7 @@ typedef enum { ncclSuccess = 0,
|
||||
|
||||
/* Generates a unique Id with each call. Used to generate commId for
|
||||
* ncclCommInitAll. uniqueId will be created in such a way that it is
|
||||
* guaranteed to be unique across the host. */
|
||||
* guaranteed to be unique accross the host. */
|
||||
ncclResult_t ncclGetUniqueId(ncclUniqueId* uniqueId);
|
||||
ncclResult_t pncclGetUniqueId(ncclUniqueId* uniqueId);
|
||||
|
||||
|
||||
@ -23,7 +23,7 @@
|
||||
* PostFlag. The primitive routines wait for all WaitFlag args to attain
|
||||
* at least a value of SUBSTEPS*(step-1)+substep+1 (i.e. completion of
|
||||
* corresponding substep by previous step) before executing the transfer.
|
||||
* After each substep is transferred, all PostFlag arguments get updated to
|
||||
* After each substep is transfered, all PostFlag arguments get updated to
|
||||
* the value SUBSTEPS*step+substep+1.
|
||||
*/
|
||||
|
||||
|
||||
@ -10,7 +10,7 @@ import torch.utils.hooks as hooks
|
||||
|
||||
def _addindent(s_, numSpaces):
|
||||
s = s_.split('\n')
|
||||
# don't do anything for single-line stuff
|
||||
# dont do anything for single-line stuff
|
||||
if len(s) == 1:
|
||||
return s_
|
||||
first = s.pop(0)
|
||||
|
||||
@ -6,7 +6,7 @@ from ._functions import Scatter, Gather
|
||||
def scatter(inputs, target_gpus, dim=0):
|
||||
"""
|
||||
Slices variables into approximately equal chunks and
|
||||
distributes them across given GPUs. Duplicates
|
||||
distributes them accross given GPUs. Duplicates
|
||||
references to objects that are not variables. Does not
|
||||
support Tensors.
|
||||
"""
|
||||
|
||||
Reference in New Issue
Block a user