OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[BE][3/6] fix typos in test/ (#157637)

Browse Source 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/157637
Approved by: https://github.com/yewentao256, https://github.com/albanD
ghstack dependencies: #156605
This commit is contained in:
Xuehai Pan 2025-07-17 14:55:14 +08:00 committed by PyTorch MergeBot
parent 3f8e2e91ad
commit c8d43cbc6e
31 changed files with 77 additions and 67 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.lintrunner.toml
View File

@ -1168,7 +1168,6 @@ exclude_patterns = [
'aten/src/ATen/native/[a-pA-P]*/**',
'aten/src/ATen/[a-mA-M]*/**',
'test/**',
'test/[a-hA-h]*/**',
]
init_command = [
'python3',

2
test/ao/sparsity/test_activation_sparsifier.py
View File

@ -50,7 +50,7 @@ class TestActivationSparsifier(TestCase):
sparsifier_defaults = activation_sparsifier.defaults
combined_defaults = {**defaults, "sparse_config": sparse_config}
# more keys are populated in activation sparsifier (eventhough they may be None)
# more keys are populated in activation sparsifier (even though they may be None)
assert len(combined_defaults) <= len(activation_sparsifier.defaults)
for key, config in sparsifier_defaults.items():

4
test/ao/sparsity/test_data_sparsifier.py
View File

@ -265,7 +265,7 @@ class _BaseDataSparsiferTestCase(TestCase):
class _NormDataSparsifierTestCase(_BaseDataSparsiferTestCase):
r"""This helper test class takes in any supported type of and runs some tests.
This inherits the TestBaseDataSparsifierRuner wherein some functions are
over-ridden to take accomodate the specific sparsifier.
over-ridden to take accommodate the specific sparsifier.
TODO: Change the structure by creating a separate test case class for each
member function
"""
@ -770,7 +770,7 @@ class TestQuantizationUtils(TestCase):
# higher threshold as quantization occurs before sparsity
threshold = (
1 # zero points seem to have higher magnitude with sparsity occuring after
1 # zero points seem to have higher magnitude with sparsity occurring after
)
sl_emb1 = (torch.abs(dequant_emb1) < threshold).float().mean()

2
test/ao/sparsity/test_scheduler.py
View File

@ -188,7 +188,7 @@ class TestCubicScheduler(TestCase):
self.assertEqual(
self._get_sparsity_levels(sparsifier),
self.sorted_sparse_levels,
msg="Sparsity level is not reaching the target level afer delta_t * n steps ",
msg="Sparsity level is not reaching the target level after delta_t * n steps ",
)

6
test/benchmark_utils/test_benchmark_utils.py
View File

@ -699,14 +699,16 @@ class TestBenchmarkUtils(TestCase):
8959166 /tmp/build/80754af9/python_15996 ... a3/envs/throwaway/bin/python3.6]
...
92821 /tmp/build/80754af9/python_15996 ... a3/envs/throwaway/bin/python3.6]
91000 build/../torch/csrc/tensor/pytho ... ch/torch/lib/libtorch_python.so]
91000 build/../torch/csrc/tensor/pytho ... ch/torch/lib/libtorch_python.so] # codespell:ignore
91000 /data/users/test_user/repos/pyto ... nsors::get_default_scalar_type()
90090 ???:pthread_mutex_lock [/usr/lib64/libpthread-2.28.so]
90000 build/../c10/core/TensorImpl.h:c ... ch/torch/lib/libtorch_python.so]
90000 build/../aten/src/ATen/record_fu ... torch/torch/lib/libtorch_cpu.so]
90000 /data/users/test_user/repos/pyto ... uard(std::optional<c10::Device>)
90000 /data/users/test_user/repos/pyto ... ersionCounter::~VersionCounter()
88000 /data/users/test_user/repos/pyto ... ratorKernel*, at::Tensor const&)""",
88000 /data/users/test_user/repos/pyto ... ratorKernel*, at::Tensor const&)""".replace(
" # codespell:ignore", ""
),
)
self.regularizeAndAssertExpectedInline(

2
test/cpp/api/transformer.cpp
View File

@ -73,7 +73,7 @@ void transformer_encoder_layer_test_helper(
ASSERT_TRUE(
torch::allclose(result, ref_output, 1e-7, 1e-5, /*equal_nan=*/true));
// all 0 values are NOT masked. This should't mask anything
// all 0 values are NOT masked. This shouldn't mask anything
torch::Tensor mask = torch::tensor({{0}}, tensor_options) == 1;
result = model(
encoder_input,

2
test/cpp/jit/CMakeLists.txt
View File

@ -17,7 +17,7 @@ set(BACKEND_WITH_COMPILER_SRCS
)
if(USE_KINETO)
# Testing edge profiler for backend use
# profiler_edge should only be aded when USE_KINETO flag is on
# profiler_edge should only be added when USE_KINETO flag is on
list(APPEND BACKEND_WITH_COMPILER_SRCS
${TORCH_SRC_DIR}/csrc/jit/mobile/profiler_edge.cpp)
endif()

2
test/cpp/jit/test_backend.cpp
View File

@ -789,7 +789,7 @@ TEST(
c._save_for_mobile(ss, ExtraFilesMap(), true);
auto c_loaded = _load_for_mobile(ss);
/*
* Erro stack trace will look like this:
* Error stack trace will look like this:
* Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
* Traceback of TorchScript (most recent call last):
* File "<string>", line 3, in FunctionName_UNKNOWN

2
test/cpp/jit/test_backend_compiler_lib.cpp
View File

@ -79,7 +79,7 @@ class BackendWithCompiler : public PyTorchBackendInterface {
// forwards everything along. In a non toy setup this could grab information
// from that runtime that might be relevant to execute, such as build flags
// the resolution of the devices camera, or basically any runtime specific
// information that wouldnt be available server side where preprocess is
// information that wouldn't be available server side where preprocess is
// called.
c10::impl::GenericDict compile(
c10::IValue processed,

2
test/cpp/jit/test_lite_trainer.cpp
View File

@ -78,7 +78,7 @@ TEST(LiteTrainerTest, Params) {
AT_ASSERT(parameters[0].item<float>() == bc_parameters[0].item<float>());
}
// TODO Renable these tests after parameters are correctly loaded on mobile
// TODO Re-enable these tests after parameters are correctly loaded on mobile
/*
TEST(MobileTest, NamedParameters) {
Module m("m");

2
test/cpp/lite_interpreter_runtime/test_lite_interpreter_runtime.cpp
View File

@ -106,7 +106,7 @@ TEST(RunTimeTest, DelegateException) {
* inputs.emplace_back(torch::rand({2, 4}));
* inputs.emplace_back(torch::rand({13, 9}));
* Run with inputs and expect exception
* Erro stack trace will look like this:
* Error stack trace will look like this:
* Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
* Traceback of TorchScript (most recent call last):
* File "<string>", line 3, in FunctionName_UNKNOWN

2
test/cpp/tensorexpr/test_cuda.cpp
View File

@ -1681,7 +1681,7 @@ TEST(Cuda, MaskMultiDim_CUDA) {
// Tests the case where loop extents are symbolic and not known at compile time.
// In this case both stores must be masked against the extent of the other loop,
// incase it is larger.
// in case it is larger.
TEST(Cuda, MaskMultiDimSymbolic_CUDA) {
VarHandle OUTER_SIZE("OUTER_SIZE", kLong);
VarHandle A_SIZE("A_SIZE", kLong);

10
test/cpp/tensorexpr/test_kernel.cpp
View File

@ -1113,7 +1113,7 @@ TEST_F(Kernel, Softmax2D) {
const auto verification_pattern =
format(verification_template, ver_env);
// verification sting temporarily disabled until
// verification string temporarily disabled until
// inlining of exp() is benchmarked and determined
// torch::jit::testing::FileCheck().run(verification_pattern,
// oss.str());
@ -1192,7 +1192,7 @@ TEST_F(Kernel, Softmax3D) {
ver_env.d("softmax_dim_size", softmax_dim_size);
const auto verification_pattern = format(verification_template, ver_env);
// verification sting temporarily disabled until
// verification string temporarily disabled until
// inlining of exp() is benchmarked and determined
// torch::jit::testing::FileCheck().run(verification_pattern, oss.str());
@ -1275,7 +1275,7 @@ TEST_F(Kernel, Softmax4D) {
ver_env.d("softmax_dim_size", softmax_dim_size);
const auto verification_pattern = format(verification_template, ver_env);
// verification sting temporarily disabled until
// verification string temporarily disabled until
// inlining of exp() is benchmarked and determined
// torch::jit::testing::FileCheck().run(verification_pattern, oss.str());
@ -1887,7 +1887,7 @@ graph(%x : int,
auto rt = at::zeros({3}, TensorOptions(kCPU).dtype(at::kLong));
// Verify that TEK::runFast works correctly with mixed scalar and tensor
// inputs/utputs
// inputs/outputs
std::vector<void*> inputs = {&x, xt.data_ptr(), &y, yt.data_ptr()};
std::vector<void*> outputs = {&r, rt.data_ptr(), &z, zt.data_ptr()};
k.runFast(inputs, outputs);
@ -1897,7 +1897,7 @@ graph(%x : int,
ASSERT_TRUE(at::equal(rt, zt * xt));
// Verify that TEK::run works correctly with mixed scalar and tensor
// inputs/utputs
// inputs/outputs
std::vector<IValue> stack = {x, xt, y, yt};
k.run(stack);
TORCH_CHECK_EQ(stack[0], x * y * x);

8
test/cpp/tensorexpr/test_memdependency.cpp
View File

@ -437,12 +437,12 @@ TEST(MemDependency, BoundSubtractMultiDim) {
ASSERT_TRUE(EQ(
subtractIndicesBounds({CB(0, 9), CB(0, 2)}, {CB(0, 10), CB(0, 20)}), {}));
// Mutli dim one way partial in dim 1.
// Multi dim one way partial in dim 1.
ASSERT_TRUE(
EQ(subtractIndicesBounds({CB(0, 9), CB(0, 2)}, {CB(0, 3), CB(0, 2)}),
{{CB(4, 9), CB(0, 2)}}));
// Mutli dim one way partial in dim 2.
// Multi dim one way partial in dim 2.
ASSERT_TRUE(
EQ(subtractIndicesBounds({CB(0, 9), CB(0, 20)}, {CB(0, 9), CB(0, 10)}),
{{CB(0, 9), CB(11, 20)}}));
@ -939,7 +939,7 @@ TEST(MemDependency, MemDependencyCheckerLoopBounds) {
*/
// Now let's look at the bounds of each access.
// There are 9 accesses in this Stmt, so this is exhaustive, we wont do this
// There are 9 accesses in this Stmt, so this is exhaustive, we won't do this
// much.
auto history = analyzer.getHistory();
ASSERT_EQ(history.size(), 10);
@ -1134,7 +1134,7 @@ TEST(MemDependency, MemDependencyCheckerLoopBoundsIndexShift) {
// this case -1.
ASSERT_TRUE(EQ(history[1]->bounds(), {CB(0, 8)}));
// It depends on the input, but also the store in the same loop, since
// different interations of the loop depend on each other.
// different iterations of the loop depend on each other.
ASSERT_EQ(history[1]->dependencies().size(), 2);
ASSERT_TRUE(history[1]->hasDependency(history[0]));
ASSERT_TRUE(history[1]->hasDependency(history[2]));

2
test/cpp/tensorexpr/test_reductions.cpp
View File

@ -333,7 +333,7 @@ TEST(Reductions, ReduceMinCustomInitializer) {
cg.call({in, out, std::numeric_limits<float>::max()});
ASSERT_EQ(out[0], 10);
// With an initalizer lower than the min, that's the min.
// With an initializer lower than the min, that's the min.
cg.call({in, out, 5.f});
ASSERT_EQ(out[0], 5);
}

8
test/cpp/tensorexpr/test_registerizer.cpp
View File

@ -1254,7 +1254,7 @@ TEST(Registerizer, RegisterizerConditionInsideOverlap2) {
* A[0] = 3;
* A[x] = (A[x]) + 1;
* }
* int A_2 = A[x]; // A_2 initialier
* int A_2 = A[x]; // A_2 initializer
* B[x] = A_2; //
* B[x + 1] = A_2; //
* A_2 = C[x]; //
@ -3064,7 +3064,7 @@ TEST(Registerizer, RegisterizerHiddenAccessNo) {
}
// In this case the conditional access must be hoisted by two loops, there are
// two accesses here one is unhidden and the other isnt. A[0] can be
// two accesses here one is unhidden and the other isn't. A[0] can be
// registerized but B[0] cannot.
TEST(Registerizer, RegisterizerHiddenAccessMultiLoop) {
BufHandle a("A", {10}, kInt);
@ -3422,8 +3422,8 @@ TEST(Registerizer, RegisterizerMultiDim) {
torch::jit::testing::FileCheck().run(verification_pattern, oss.str());
}
// Wont registerize if only some dims match, but will still registerize distinct
// elements.
// Won't registerize if only some dims match, but will still registerize
// distinct elements.
TEST(Registerizer, RegisterizerMultiDimPartial) {
BufHandle a("A", {3, 4, 5}, kInt);
VarHandle x("x", kInt);

16
test/cpp/tensorexpr/test_simplify.cpp
View File

@ -2643,7 +2643,7 @@ TEST(Simplify, SimplifyWontReorderFloat) {
VarHandle x("x", kFloat);
VarHandle y("y", kFloat);
// x%y - (x%y - 1) => x%y - (x%y - 1).
// We wont reorder opaque ops if they are FP.
// We won't reorder opaque ops if they are FP.
ExprHandle body = (x % y) - ((x % y) - 1);
ExprHandle simplified = IRSimplifier::simplify(body);
@ -2794,7 +2794,7 @@ TEST(Simplify, SimplifyRoundModPattern) {
}
{
// Sanity checking we wont do the optimization on floats.
// Sanity checking we won't do the optimization on floats.
VarHandle x("x", kFloat);
VarHandle y("y", kFloat);
ExprHandle body = ((x / y) * y) + (x % y);
@ -2811,7 +2811,7 @@ TEST(Simplify, SimplifyRoundModPattern) {
}
{
// Sanity check we wont do it if the mod term doesn't match.
// Sanity check we won't do it if the mod term doesn't match.
VarHandle x("x", kInt);
VarHandle y("y", kInt);
VarHandle z("z", kInt);
@ -2821,7 +2821,7 @@ TEST(Simplify, SimplifyRoundModPattern) {
}
{
// Sanity check we wont do it if the div term doesn't match.
// Sanity check we won't do it if the div term doesn't match.
VarHandle x("x", kInt);
VarHandle y("y", kInt);
VarHandle z("z", kInt);
@ -2831,7 +2831,7 @@ TEST(Simplify, SimplifyRoundModPattern) {
}
{
// Sanity check we wont do it if the mul term doesn't match.
// Sanity check we won't do it if the mul term doesn't match.
VarHandle x("x", kInt);
VarHandle y("y", kInt);
VarHandle z("z", kInt);
@ -3013,7 +3013,7 @@ TEST(Simplify, SimplifyModRoundModPattern) {
}
{
// Sanity checking we wont do the optimization on floats.
// Sanity checking we won't do the optimization on floats.
VarHandle x("x", kFloat);
VarHandle y("y", kFloat);
VarHandle z("z", kFloat);
@ -4264,7 +4264,7 @@ TEST(Simplify, SimplifyReorderForCond) {
{
// Condition uses distinct region of Tensor.
// We could reorder here wih better analysis, but we don't. Included for
// We could reorder here with better analysis, but we don't. Included for
// completeness.
auto body = For::make(
i,
@ -4643,7 +4643,7 @@ TEST(Simplify, SimplifyFuseConditions) {
}
{
// Sanity check wont fuse different non-CompareSelects.
// Sanity check won't fuse different non-CompareSelects.
auto body = Block::make(
{Cond::make(i, Store::make(a, {0}, i), nullptr),
Cond::make(j, Store::make(a, {1}, i), nullptr)});

6
test/export/test_converter.py
View File

@ -700,7 +700,7 @@ class TestConverter(TestCase):
else:
return self.w + self.m2(x)
# Super nested, parameters neeed to lifted
# Super nested, parameters need to be lifted
# multiple times.
class SuperNestedM(torch.nn.Module):
def __init__(self) -> None:
@ -755,7 +755,7 @@ class TestConverter(TestCase):
else:
return self.linear(self.m2(x))
# Super nested, parameters neeed to lifted
# Super nested, parameters need to be lifted
# multiple times.
class SuperNestedM1(torch.nn.Module):
def __init__(self, dim: int) -> None:
@ -771,7 +771,7 @@ class TestConverter(TestCase):
return self.linear(self.m2(x))
# Super nested, even the input needs to be
# lifted recursively due to value propogation optimiztaion.
# lifted recursively due to value propagation optimization.
class SuperNestedM2(torch.nn.Module):
def __init__(self, dim: int) -> None:
super().__init__()

8
test/export/test_export.py
View File

@ -1871,7 +1871,7 @@ class GraphModule(torch.nn.Module):
for problem in [Problem1, Problem2]:
m = problem()
m(torch.rand(64, 64))
# simpified torch.distributed.pipeline code
# simplified torch.distributed.pipeline code
annotate_split_points(m, {"blocks.1": 1, "blocks.3": 1})
gm = export(m, (torch.rand(64, 64),))
torch.export.unflatten(gm)
@ -8096,7 +8096,7 @@ def forward(self, x):
str(schema),
"""cond(SymBool pred, GraphModule true_fn, GraphModule false_fn, Tensor[2] operands) -> Tensor[1]""",
)
# serdes deserailizes tuple as list
# serdes deserializes tuple as list
if need_serdes_test(self._testMethodName):
self.assertExpectedInline(
ep.graph_module.code.strip(),
@ -9232,7 +9232,7 @@ graph():
x = torch.rand(5, 2, 2)
model = Model()
# Manualy set the fake_device of fake tensors.
# Manually set the fake_device of fake tensors.
x.fake_device = torch.device("cuda:0")
for n, p in model.named_parameters():
p.fake_device = torch.device("cuda:0")
@ -13562,7 +13562,7 @@ graph():
self.assertTrue(torch.allclose(m(x2), ep.module()(x2)))
self.assertTrue(torch.allclose(m(x1), ep.module()(x1)))
@testing.expectedFailureSerDerNonStrict # construtor is not serialized today
@testing.expectedFailureSerDerNonStrict # constructor is not serialized today
@testing.expectedFailureSerDer # constructor is not serialized today
@testing.expectedFailureRetraceability # dynamo doesn't work with FlatApply op
def test_capture_subclass_constructor(self):

2
test/export/test_torchbind.py
View File

@ -1274,7 +1274,7 @@ class TestCompileTorchbind(TestCase):
self.assertEqual(cnt.frame_count, 1)
tq2 = _empty_tensor_queue()
# make first tensor's secon dim dynamic
# make first tensor's second dim dynamic
tq2.push(torch.randn(2, 4, requires_grad=False))
torch.compile(mod, backend=cnt)(tq2, x)
self.assertEqual(cnt.frame_count, 2)

2
test/functorch/attn_ft.py
View File

@ -126,7 +126,7 @@ class BertSelfAttention(nn.Module):
if self.position_embedding_type == "relative_key":
# these were einsum ops in the positional code because they are not easy to fit to existing matmul operators
# eventhough they are degenerate matmuls
# even though they are degenerate matmuls
relative_position_scores = (q * positional_embedding).sum(features)
attention_scores = attention_scores + relative_position_scores
elif self.position_embedding_type == "relative_key_query":

6
test/functorch/test_aotdispatch.py
View File

@ -2372,7 +2372,7 @@ def forward(self, arg0_1, arg1_1):
return a.mul(3), b.mul(4)
inp = [
# First inp doesnt require grad, but we switch it on
# First inp doesn't require grad, but we switch it on
torch.ones(3, 3, requires_grad=False),
torch.ones(3, 3, requires_grad=True),
]
@ -5670,7 +5670,7 @@ def forward(self, primals_1, tangents_1):
_, fw_graph_out_nodes = get_ins_outs(fw_graph)
self.assertEqual(
# fw outputs include b.size() which expands to 2 symints,
# then 4 tensors (transposes of matricies used for mm) are saved
# then 4 tensors (transposes of matrices used for mm) are saved
# finally 3 symints are saved
[False, True, True, False, False] + [False] * 4 + [True] * 3,
[is_sym_node(n) for n in fw_graph_out_nodes],
@ -6000,7 +6000,7 @@ metadata incorrectly.
self.assertEqual(b_test.a, b_ref.a)
self.assertEqual(b_test.b, b_ref.b)
# NOTE: we need to use b in our gradient compute. Otherwise we will need to recompile teh backward.
# NOTE: we need to use b in our gradient compute. Otherwise we will need to recompile the backward.
(b_ref * out_ref).sum().backward()
(b_test * out_test).sum().backward()
# Both grad_inputs are TwoTensor

10
test/functorch/test_control_flow.py
View File

@ -2013,7 +2013,7 @@ def forward(self, pred_1, x_1):
if autograd:
self.check_autograd(result, expected_result, (init, inp))
# TODO: Does not work because of the usage of vmap witin associative_scan
# TODO: Does not work because of the usage of vmap within associative_scan
# The paT206899919 rameterization is commented out for the moment and the test is marked with expected fail
# Fails with: AssertionError: scan is not an OpOverload
@skipIfRocm(msg="Unsupported on ROCM yet")
@ -4143,7 +4143,7 @@ class GraphModule(torch.nn.Module):
inputs=inp,
)
# TODO: Does not work because of the usage of vmap witin associative_scan
# TODO: Does not work because of the usage of vmap within associative_scan
# TODO: Re-enable additional parameters again once this issues has been resolved
@unittest.skipIf(not SM70OrLater, "triton")
@requires_cuda
@ -4241,7 +4241,7 @@ class GraphModule(torch.nn.Module):
inputs=inp,
)
# TODO: Does not work because of the usage of vmap witin associative_scan
# TODO: Does not work because of the usage of vmap within associative_scan
# TODO: Re-enable additional parameters again once this issues has been resolved
@unittest.skipIf(not SM70OrLater, "triton")
@requires_cuda
@ -4314,7 +4314,7 @@ class GraphModule(torch.nn.Module):
inputs=inp,
)
# TODO: Does not work because of the usage of vmap witin associative_scan
# TODO: Does not work because of the usage of vmap within associative_scan
# TODO: Re-enable additional parameters again once this issues has been resolved
@unittest.skipIf(not SM70OrLater, "triton")
@requires_cuda
@ -5315,7 +5315,7 @@ def forward(self, arg0_1):
)
@parametrize("func_type", ["no", "cpp", "python", "functorch"])
# - "simple_with_linear" and "nested_with_linear" doesn't work becaue parameters and buffers
# - "simple_with_linear" and "nested_with_linear" doesn't work because parameters and buffers
# are not inputs so they're not wrapped by functionalization and tracing.
#
# - make_fx tracing mode "real" fails for "int_carry", "pytree_int_carry" and "const_and_symint_output"

4
test/functorch/test_vmap.py
View File

@ -4152,7 +4152,7 @@ class TestVmapOperatorsOpInfo(TestCase):
with subtest_ctx(self), skip_xfail_ctx(self):
args = (sample_input.input,) + sample_input.args
if not any(isinstance(arg, torch.Tensor) for arg in args):
# Atleast one tensor required for vmap.
# At least one tensor required for vmap.
continue
kwargs = sample_input.kwargs
is_batch_norm_and_training = is_batch_norm_training(op.name, kwargs)
@ -4230,7 +4230,7 @@ class TestVmapOperatorsOpInfo(TestCase):
xfail("as_strided_copy"),
xfail(
"as_strided_scatter"
), # no batching rule implemented, default doesnt work
), # no batching rule implemented, default doesn't work
skip(
"new_empty_strided"
), # empty tensor data is garbage so it's hard to make comparisons with it

6
test/fx/test_lazy_graph_module.py
View File

@ -69,7 +69,7 @@ class TestLazyGraphModule(TestCase):
def test_needs_recompile(self):
"""
Make sure needs_recompile() return the corrent state.
Make sure needs_recompile() return the correct state.
"""
def f(x):
@ -141,7 +141,7 @@ class TestLazyGraphModule(TestCase):
self.assertTrue(isinstance(gm2, _LazyGraphModule))
self.assertTrue(gm2._needs_recompile())
# make_fx will cal foward method of gm. That clears the _needs_recompile()
# make_fx will cal forward method of gm. That clears the _needs_recompile()
# flag.
self.assertFalse(gm._needs_recompile())
@ -175,7 +175,7 @@ class TestLazyGraphModule(TestCase):
def test_save_lazy_foward(self):
"""
Save the lazy forward method and call it repeatly. Make sure we
Save the lazy forward method and call it repeatedly. Make sure we
don't recompile for each such call.
"""

6
test/fx/test_partitioner_order.py
View File

@ -36,9 +36,9 @@ class TestPartitionerOrder(TestCase):
def test_partitioner_order(self):
m = AddModule()
traced_m = torch.fx.symbolic_trace(m)
partions = DummyPartitioner(traced_m).propose_partitions()
partion_nodes = [list(partition.nodes) for partition in partions]
node_order = [n.name for n in partion_nodes[0]]
partitions = DummyPartitioner(traced_m).propose_partitions()
partition_nodes = [list(partition.nodes) for partition in partitions]
node_order = [n.name for n in partition_nodes[0]]
for _ in range(10):
traced_m = torch.fx.symbolic_trace(m)
new_partion = DummyPartitioner(traced_m).propose_partitions()

2
test/fx/test_pass_infra.py
View File

@ -131,7 +131,7 @@ class TestPassManager(TestCase):
def test_topological_sort(self):
"""
Tests that passes are correctly ordered based on contraints.
Tests that passes are correctly ordered based on constraints.
"""
def pass0(x):

2
test/higher_order_ops/test_invoke_quant.py
View File

@ -186,7 +186,7 @@ class TestInvokeQuantInductor(TestInvokeQuant):
@skipIfXpu(
msg="MM Triton template fusion for XPU not work because the fusion"
" can not speedup, unskip untill #146568 fixed."
" can not speedup, unskip until #146568 fixed."
)
@requires_gpu()
@config.patch(prologue_fusion=True)

4
test/higher_order_ops/test_invoke_subgraph.py
View File

@ -1079,7 +1079,7 @@ class GraphModule(torch.nn.Module):
fake_prop_count = 0
def _mock_invoke_subgraph(mode, subgraph, identifer, *operands):
def _mock_invoke_subgraph(mode, subgraph, identifier, *operands):
nonlocal fake_prop_count
fake_prop_count += 1
return (operands[0].clone(),)
@ -2077,7 +2077,7 @@ class GraphModule(torch.nn.Module):
# NOTE THAT THIS TEST DOES NOT REALLY WORK
# We wanted one invoke_subgraph called twice, but because of
# constant_args_idx changing in the grpah, the graph equivalence fails
# constant_args_idx changing in the graph, the graph equivalence fails
if not TEST_WITH_CROSSREF:
self.assertExpectedInline(

9
tools/linter/dictionary.txt
View File

@ -1,9 +1,18 @@
aLoad
aLoads
ans
aStore
aStores
belows
bLoad
bLoads
bStore
bStores
BU
contiguities
contiguity
coo
deser
din
dout
ElementE

4
torch/csrc/jit/tensorexpr/loopnest.cpp
View File

@ -1843,11 +1843,11 @@ bool LoopNest::hasLoopCarriedDependence(const ForPtr& loop) {
auto bLoads = NodeFinder<Load>::find(*it2);
// ReadAfterWrite
for (auto& aStore : aStores) {
for (auto& bLoad : bLoads) { // codespell:ignore
for (auto& bLoad : bLoads) {
if (aStore->buf() == bLoad->buf()) {
if (!areIndicesLoopIndependent(
aStore->indices(), bLoad->indices(), outer_loop_vars)) {
if (isOverlapping(analyzer, aStore, bLoad)) { // codespell:ignore
if (isOverlapping(analyzer, aStore, bLoad)) {
return true;
}
}