Add missing skipIf(not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION)

.ci/docker/requirements-docs.txt

@@ -1,11 +1,15 @@
-sphinx==7.2.6
+sphinx==5.3.0
 #Description: This is used to generate PyTorch docs
-#Pinned versions: 7.2.6
+#Pinned versions: 5.3.0
 
-pytorch_sphinx_theme2==0.2.0
-#Description: This is needed to generate PyTorch docs
-#Pinned versions: 0.2.0
+standard-imghdr==3.13.0; python_version >= "3.13"
+#Description: This is needed by Sphinx, so it needs to be added here.
+# The reasons are as follows:
+# 1) This module has been removed from the Python standard library since Python 3.13(https://peps.python.org/pep-0594/#imghdr);
+# 2) The current version of Sphinx (5.3.0) is not compatible with Python 3.13.
+# Once Sphinx is upgraded to a version compatible with Python 3.13 or later, we can remove this dependency.
 
+-e git+https://github.com/pytorch/pytorch_sphinx_theme.git@71e55749be14ceb56e7f8211a9fb649866b87ad4#egg=pytorch_sphinx_theme2
 # TODO: sphinxcontrib.katex 0.9.0 adds a local KaTeX server to speed up pre-rendering
 # but it doesn't seem to work and hangs around idly. The initial thought that it is probably
 # something related to Docker setup. We can investigate this later.
@@ -32,17 +36,17 @@ tensorboard==2.18.0 ; python_version >= "3.13"
 #Description: This is used to generate PyTorch docs
 #Pinned versions: 2.13.0
 
-breathe==4.36.0
+breathe==4.34.0
 #Description: This is used to generate PyTorch C++ docs
-#Pinned versions: 4.36.0
+#Pinned versions: 4.34.0
 
-exhale==0.3.7
+exhale==0.2.3
 #Description: This is used to generate PyTorch C++ docs
-#Pinned versions: 0.3.7
+#Pinned versions: 0.2.3
 
-docutils==0.20
+docutils==0.16
 #Description: This is used to generate PyTorch C++ docs
-#Pinned versions: 0.20
+#Pinned versions: 0.16
 
 bs4==0.0.1
 #Description: This is used to generate PyTorch C++ docs
@@ -52,13 +56,13 @@ IPython==8.12.0
 #Description: This is used to generate PyTorch functorch docs
 #Pinned versions: 8.12.0
 
-myst-nb==1.3.0
+myst-nb==0.17.2
 #Description: This is used to generate PyTorch functorch and torch.compile docs.
-#Pinned versions: 1.3.0
+#Pinned versions: 0.17.2
 
 # The following are required to build torch.distributed.elastic.rendezvous.etcd* docs
 python-etcd==0.4.5
 sphinx-copybutton==0.5.0
-sphinx-design==0.6.1
+sphinx-design==0.4.0
 sphinxcontrib-mermaid==1.0.0
-myst-parser==4.0.1
+myst-parser==0.18.1

.ci/pytorch/python_doc_push_script.sh

@@ -89,41 +89,23 @@ if [ "$is_main_doc" = true ]; then
 
   make coverage
   # Now we have the coverage report, we need to make sure it is empty.
-  # Sphinx 7.2.6+ format: python.txt contains a statistics table with a TOTAL row
-  # showing the undocumented count in the third column.
-  # Example: | TOTAL | 99.83% | 2 |
+  # Count the number of lines in the file and turn that number into a variable
+  # $lines. The `cut -f1 ...` is to only parse the number, not the filename
+  # Skip the report header by subtracting 2: the header will be output even if
+  # there are no undocumented items.
   #
   # Also: see docs/source/conf.py for "coverage_ignore*" items, which should
   # be documented then removed from there.
-
-  # Extract undocumented count from TOTAL row in Sphinx 7.2.6 statistics table
-  # The table format is: | Module | Coverage | Undocumented |
-  # Extract the third column (undocumented count) from the TOTAL row
-  undocumented=$(grep "| TOTAL" build/coverage/python.txt | awk -F'|' '{print $4}' | tr -d ' ')
-
-  if [ -z "$undocumented" ] || ! [[ "$undocumented" =~ ^[0-9]+$ ]]; then
+  lines=$(wc -l build/coverage/python.txt 2>/dev/null |cut -f1 -d' ')
+  undocumented=$((lines - 2))
+  if [ $undocumented -lt 0 ]; then
     echo coverage output not found
     exit 1
-  elif [ "$undocumented" -gt 0 ]; then
-    set +x  # Disable command echoing for cleaner output
-    echo ""
-    echo "====================="
-    echo "UNDOCUMENTED OBJECTS:"
-    echo "====================="
-    echo ""
-    # Find the line number of the TOTAL row and print only what comes after it
-    total_line=$(grep -n "| TOTAL" build/coverage/python.txt | cut -d: -f1)
-    if [ -n "$total_line" ]; then
-      # Print only the detailed list (skip the statistics table)
-      tail -n +$((total_line + 2)) build/coverage/python.txt
-    else
-      # Fallback to showing entire file if TOTAL line not found
-      cat build/coverage/python.txt
-    fi
-    echo ""
+  elif [ $undocumented -gt 0 ]; then
+    echo undocumented objects found:
+    cat build/coverage/python.txt
     echo "Make sure you've updated relevant .rsts in docs/source!"
-    echo "You can reproduce locally by running 'cd docs && make coverage && tail -n +\$((grep -n \"| TOTAL\" build/coverage/python.txt | cut -d: -f1) + 2)) build/coverage/python.txt'"
-    set -x  # Re-enable command echoing
+    echo "You can reproduce locally by running 'cd docs && make coverage && cat build/coverage/python.txt'"
     exit 1
   fi
 else

SECURITY.md

@@ -1,7 +1,7 @@
 # Security Policy
 
  - [**Reporting a Vulnerability**](#reporting-a-vulnerability)
- - [**Using PyTorch Securely**](#using-pytorch-securely)
+ - [**Using Pytorch Securely**](#using-pytorch-securely)
    - [Untrusted models](#untrusted-models)
    - [TorchScript models](#torchscript-models)
    - [Untrusted inputs](#untrusted-inputs)
@@ -10,28 +10,28 @@
 - [**CI/CD security principles**](#cicd-security-principles)
 ## Reporting Security Issues
 
-Beware that none of the topics under [Using PyTorch Securely](#using-pytorch-securely) are considered vulnerabilities of PyTorch.
+Beware that none of the topics under [Using Pytorch Securely](#using-pytorch-securely) are considered vulnerabilities of Pytorch.
 
 However, if you believe you have found a security vulnerability in PyTorch, we encourage you to let us know right away. We will investigate all legitimate reports and do our best to quickly fix the problem.
 
 Please report security issues using https://github.com/pytorch/pytorch/security/advisories/new
 
-All reports submitted through the security advisories mechanism would **either be made public or dismissed by the team within 90 days of the submission**. If advisory has been closed on the grounds that it is not a security issue, please do not hesitate to create an [new issue](https://github.com/pytorch/pytorch/issues/new?template=bug-report.yml) as it is still likely a valid issue within the framework.
+All reports submitted thru the security advisories mechanism would **either be made public or dismissed by the team within 90 days of the submission**. If advisory has been closed on the grounds that it is not a security issue, please do not hesitate to create an [new issue](https://github.com/pytorch/pytorch/issues/new?template=bug-report.yml) as it is still likely a valid issue within the framework.
 
 Please refer to the following page for our responsible disclosure policy, reward guidelines, and those things that should not be reported:
 
 https://www.facebook.com/whitehat
 
 
-## Using PyTorch Securely
-**PyTorch models are programs**, so treat its security seriously -- running untrusted models is equivalent to running untrusted code. In general we recommend that model weights and the python code for the model are distributed independently. That said, be careful about where you get the python code from and who wrote it (preferentially check for a provenance or checksums, do not run any pip installed package).
+## Using Pytorch Securely
+**Pytorch models are programs**, so treat its security seriously -- running untrusted models is equivalent to running untrusted code. In general we recommend that model weights and the python code for the model are distributed independently. That said, be careful about where you get the python code from and who wrote it (preferentially check for a provenance or checksums, do not run any pip installed package).
 
 ### Untrusted models
 Be careful when running untrusted models. This classification includes models created by unknown developers or utilizing data obtained from unknown sources[^data-poisoning-sources].
 
 **Prefer to execute untrusted models within a secure, isolated environment such as a sandbox** (e.g., containers, virtual machines). This helps protect your system from potentially malicious code. You can find further details and instructions in [this page](https://developers.google.com/code-sandboxing).
 
-**Be mindful of risky model formats**. Give preference to share and load weights with the appropriate format for your use case. [Safetensors](https://huggingface.co/docs/safetensors/en/index) gives the most safety but is the most restricted in what it supports. [`torch.load`](https://pytorch.org/docs/stable/generated/torch.load.html#torch.load) has a significantly larger surface of attack but is more flexible in what it can serialize. See the documentation for more details.
+**Be mindful of risky model formats**. Give preference to share and load weights with the appropriate format for your use case. [safetensors](https://huggingface.co/docs/safetensors/en/index) gives the most safety but is the most restricted in what it supports. [`torch.load`](https://pytorch.org/docs/stable/generated/torch.load.html#torch.load) has a significantly larger surface of attack but is more flexible in what it can serialize. See the documentation for more details.
 
 Even for more secure serialization formats, unexpected inputs to the downstream system can cause diverse security threats (e.g. denial of service, out of bound reads/writes) and thus we recommend extensive validation of any untrusted inputs.
 
@@ -43,7 +43,7 @@ Important Note: The trustworthiness of a model is not binary. You must always de
 
 ### TorchScript models
 
-TorchScript models should be treated the same way as locally executable code from an unknown source. Only run TorchScript models if you trust the provider. Please note, that tools for introspecting TorchScript models (such as `torch.utils.model_dump`) may also execute partial or full code stored in those models, therefore they should be used only if you trust the provider of the binary you are about to load.
+TorchScript models should treated the same way as locally executable code from an unknown source. Only run TorchScript models if you trust the provider. Please note, that tools for introspecting TorchScript models (such as `torch.utils.model_dump`) may also execute partial or full code stored in those models, therefore they should be used only if you trust the provider of the binary you are about to load.
 
 ### Untrusted inputs during training and prediction
 
@@ -59,9 +59,9 @@ If applicable, prepare your model against bad inputs and prompt injections. Some
 
 ### Data privacy
 
-**Take special security measures if you train your models with sensitive data**. Prioritize [sandboxing](https://developers.google.com/code-sandboxing) your models and:
-- Do not feed sensitive data to an untrusted model (even if runs in a sandboxed environment)
-- If you consider publishing a model that was partially trained with sensitive data, be aware that data can potentially be recovered from the trained weights (especially if the model overfits).
+**Take special security measures if your model if you train models with sensitive data**. Prioritize [sandboxing](https://developers.google.com/code-sandboxing) your models and:
+- Do not feed sensitive data to untrusted model (even if runs in a sandboxed environment)
+- If you consider publishing a model that was partially trained with sensitive data, be aware that data can potentially be recovered from the trained weights (especially if model overfits).
 
 ### Using distributed features
 

aten/src/ATen/Context.cpp

@@ -23,6 +23,8 @@ C10_DIAGNOSTIC_POP()
 #endif
 namespace at {
 
+namespace {
+
 /*
   These const variables defined the fp32 precisions for different backend
   We have "generic", "cuda", "mkldnn" backend now and we can choose fp32
@@ -39,6 +41,16 @@ namespace at {
                 ->rnn
 */
 
+  C10_ALWAYS_INLINE void warn_deprecated_fp32_precision_api(){
+    TORCH_WARN_ONCE(
+      "Please use the new API settings to control TF32 behavior, such as torch.backends.cudnn.conv.fp32_precision = 'tf32' "
+      "or torch.backends.cuda.matmul.fp32_precision = 'ieee'. Old settings, e.g, torch.backends.cuda.matmul.allow_tf32 = True, "
+      "torch.backends.cudnn.allow_tf32 = True, allowTF32CuDNN() and allowTF32CuBLAS() will be deprecated after Pytorch 2.9. Please see "
+      "https://pytorch.org/docs/main/notes/cuda.html#tensorfloat-32-tf32-on-ampere-and-later-devices"
+    );
+  }
+} // namespace
+
 Float32Backend str2backend(const std::string& name) {
   if (name == "generic")
     return Float32Backend::GENERIC;
@@ -194,6 +206,7 @@ bool Context::allowTF32CuDNN(std::optional<Float32Op> op) const {
   } else {
     return float32Precision(Float32Backend::CUDA, op.value()) == Float32Precision::TF32;
   }
+  warn_deprecated_fp32_precision_api();
   return allow_tf32_cudnn;
 }
 
@@ -201,6 +214,7 @@ void Context::setAllowTF32CuDNN(bool b) {
   setFloat32Precision(Float32Backend::CUDA, Float32Op::RNN, b ? Float32Precision::TF32 : Float32Precision::NONE);
   setFloat32Precision(Float32Backend::CUDA, Float32Op::CONV, b ? Float32Precision::TF32 : Float32Precision::NONE);
   allow_tf32_cudnn = b;
+  warn_deprecated_fp32_precision_api();
 }
 
 void Context::setSDPPriorityOrder(const std::vector<int64_t>& order) {
@@ -311,6 +325,7 @@ bool Context::allowTF32CuBLAS() const {
       "Current status indicate that you have used mix of the legacy and new APIs to set the TF32 status for cublas matmul. ",
       "We suggest only using the new API to set the TF32 flag. See also: ",
       "https://pytorch.org/docs/main/notes/cuda.html#tensorfloat-32-tf32-on-ampere-and-later-devices");
+  warn_deprecated_fp32_precision_api();
   return allow_tf32_new;
 }
 
@@ -334,6 +349,7 @@ Float32MatmulPrecision Context::float32MatmulPrecision() const {
       "Current status indicate that you have used mix of the legacy and new APIs to set the matmul precision. ",
       "We suggest only using the new API for matmul precision. See also: ",
       "https://pytorch.org/docs/main/notes/cuda.html#tensorfloat-32-tf32-on-ampere-and-later-devices");
+  warn_deprecated_fp32_precision_api();
   return float32_matmul_precision;
 }
 
@@ -361,6 +377,7 @@ Float32Precision Context::float32Precision(Float32Backend backend, Float32Op op)
 
 void Context::setFloat32MatmulPrecision(const std::string &s) {
   auto match = [this](const std::string & s_) {
+    warn_deprecated_fp32_precision_api();
     // TODO: consider if CuDNN field needs to also be set for potential future CuDNN ops like multi-headed attention
     if (s_ == "highest") {
       float32_matmul_precision = at::Float32MatmulPrecision::HIGHEST;

docs/source/conf.py

@@ -206,43 +206,19 @@ templates_path = [
     os.path.join(os.path.dirname(pytorch_sphinx_theme2.__file__), "templates"),
 ]
 # TODO: document these and remove them from here.
-# Fixes the duplicated
-autosummary_filename_map = {
-    "torch.nn.utils.prune.identity": "torch.nn.utils.prune.identity_function",
-    "torch.nn.utils.prune.Identity": "torch.nn.utils.prune.Identity_class",
-    "torch.optim.adamw.adamw": "torch.optim.adamw.adamw_function",
-    "torch.optim.adamw.AdamW": "torch.optim.adamw.AdamW_class",
-    "torch.optim.asgd.asgd": "torch.optim.asgd.asgd_function",
-    "torch.optim.asgd.ASGD": "torch.optim.asgd.ASGD_class",
-    "torch.optim.nadam.nadam": "torch.optim.nadam.nadam_function",
-    "torch.optim.nadam.NAdam": "torch.optim.nadam.NAdam_class",
-    "torch.optim.radam.radam": "torch.optim.radam.radam_function",
-    "torch.optim.radam.RAdam": "torch.optim.radam.RAdam_class",
-    "torch.optim.rmsprop.rmsprop": "torch.optim.rmsprop.rmsprop_function",
-    "torch.optim.rmsprop.RMSprop": "torch.optim.rmsprop.RMSprop_class",
-    "torch.optim.rprop.rprop": "torch.optim.rprop.rprop_function",
-    "torch.optim.rprop.Rprop": "torch.optim.rprop.Rprop_class",
-    "torch.optim.sgd.sgd": "torch.optim.sgd.sgd_function",
-    "torch.optim.sgd.SGD": "torch.optim.sgd.SGD_class",
-    "torch.optim.adadelta.adadelta": "torch.optim.adadelta.adadelta_function",
-    "torch.optim.adadelta.Adadelta": "torch.optim.adadelta.Adadelta_class",
-    "torch.optim.adagrad.adagrad": "torch.optim.adagrad.adagrad_function",
-    "torch.optim.adagrad.Adagrad": "torch.optim.adagrad.Adagrad_class",
-    "torch.optim.adam.adam": "torch.optim.adam.adam_function",
-    "torch.optim.adam.Adam": "torch.optim.adam.Adam_class",
-    "torch.optim.adamax.adamax": "torch.optim.adamax.adamax_function",
-    "torch.optim.adamax.Adamax": "torch.optim.adamax.Adamax_class",
-    "torch.mtia.stream": "torch.mtia.stream_function",
-    "torch.mtia.Stream": "torch.mtia.Stream_class",
-    "torch.cpu.stream": "torch.cpu.stream_function",
-    "torch.cpu.Stream": "torch.cpu.Stream_class",
-    "torch.cuda.stream": "torch.cuda.stream_function",
-    "torch.cuda.Stream": "torch.cuda.Stream_class",
-    "torch.xpu.stream": "torch.xpu.stream_function",
-    "torch.xpu.Stream": "torch.xpu.Stream_class",
-}
 
 coverage_ignore_functions = [
+    # torch
+    "typename",
+    # torch.cuda._sanitizer
+    "zip_arguments",
+    "zip_by_key",
+    # torch.distributed.autograd
+    "is_available",
+    # torch.distributed.checkpoint.state_dict
+    "gc_context",
+    # torch.distributed.elastic.events
+    "record_rdzv_event",
     # torch.distributed.elastic.metrics
     "initialize_metrics",
     # torch.distributed.elastic.rendezvous.registry
@@ -3219,11 +3195,6 @@ autodoc_type_aliases = {
 # Enable overriding of function signatures in the first line of the docstring.
 autodoc_docstring_signature = True
 
-# Exclude inherited IntEnum methods that have RST formatting issues in their docstrings
-autodoc_default_options = {
-    "exclude-members": "from_bytes, to_bytes",
-}
-
 # -- katex javascript in header
 #
 #    def setup(app):

docs/source/quantization-support.md

@@ -253,6 +253,7 @@ regular full-precision tensor.
 .. autosummary::
     :toctree: generated
     :nosignatures:
+    :template: classtemplate.rst
 
     view
     as_strided

test/inductor/test_compile_worker.py

@@ -4,7 +4,6 @@ import os
 import tempfile
 from threading import Event
 
-import torch._inductor.config as config
 from torch._inductor.compile_worker.subproc_pool import (
     raise_testexc,
     SubprocException,
@@ -17,12 +16,9 @@ from torch.testing._internal.inductor_utils import HAS_CPU
 
 
 class TestCompileWorker(TestCase):
-    def make_pool(self, size):
-        return SubprocPool(size)
-
     @skipIfWindows(msg="pass_fds not supported on Windows.")
     def test_basic_jobs(self):
-        pool = self.make_pool(2)
+        pool = SubprocPool(2)
         try:
             a = pool.submit(operator.add, 100, 1)
             b = pool.submit(operator.sub, 100, 1)
@@ -33,7 +29,7 @@ class TestCompileWorker(TestCase):
 
     @skipIfWindows(msg="pass_fds not supported on Windows.")
     def test_exception(self):
-        pool = self.make_pool(2)
+        pool = SubprocPool(2)
         try:
             a = pool.submit(raise_testexc)
             with self.assertRaisesRegex(
@@ -46,7 +42,7 @@ class TestCompileWorker(TestCase):
 
     @skipIfWindows(msg="pass_fds not supported on Windows.")
     def test_crash(self):
-        pool = self.make_pool(2)
+        pool = SubprocPool(2)
         try:
             with self.assertRaises(Exception):
                 a = pool.submit(os._exit, 1)
@@ -62,7 +58,7 @@ class TestCompileWorker(TestCase):
 
     @skipIfWindows(msg="pass_fds not supported on Windows.")
     def test_quiesce(self):
-        pool = self.make_pool(2)
+        pool = SubprocPool(2)
         try:
             a = pool.submit(operator.add, 100, 1)
             pool.quiesce()
@@ -79,7 +75,7 @@ class TestCompileWorker(TestCase):
         os.environ["ROLE_RANK"] = "0"
         with tempfile.NamedTemporaryFile(delete=True) as temp_log:
             os.environ["TORCHINDUCTOR_WORKER_LOGPATH"] = temp_log.name
-            pool = self.make_pool(2)
+            pool = SubprocPool(2)
             try:
                 pool.submit(operator.add, 100, 1)
                 self.assertEqual(os.path.exists(temp_log.name), True)
@@ -87,12 +83,6 @@ class TestCompileWorker(TestCase):
                 pool.shutdown()
 
 
-@config.patch("quiesce_async_compile_time", 0.1)
-class TestCompileWorkerWithTimer(TestCompileWorker):
-    def make_pool(self, size):
-        return SubprocPool(size, quiesce=True)
-
-
 class TestTimer(TestCase):
     def test_basics(self):
         done = Event()

test/test_transformers.py

@@ -1914,6 +1914,7 @@ class TestSDPAFailureModes(NNTestCase):
                     q, k, v, None, 0.0, is_causal=True))
 
     @onlyCUDA
+    @unittest.skipIf(not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Does not support Efficient Attention")
     def test_mem_eff_attention_fail_with_batch_size_geq_65536(self):
         batch_size = 2**16
         query = torch.rand([batch_size, 2, 2, 8], device='cuda', dtype=torch.float16, requires_grad=True)
@@ -1935,6 +1936,7 @@ class TestSDPAFailureModes(NNTestCase):
         self.assertEqual(value.grad, v_cpu.grad, atol=2e-3, rtol=1e-4)
 
     @onlyCUDA
+    @unittest.skipIf(not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Does not support Efficient Attention")
     def test_mem_eff_attention_fail_with_batch_size_geq_65536_error(self):
         query = torch.rand([2**16, 2, 2, 8], device='cuda', dtype=torch.float16)
         key = torch.rand([2**16, 2, 2, 8], device='cuda', dtype=torch.float16)
@@ -1948,6 +1950,7 @@ class TestSDPAFailureModes(NNTestCase):
 
     @largeTensorTest("15GB", "cuda")
     @onlyCUDA
+    @unittest.skipIf(not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Does not support Efficient Attention")
     def test_mem_eff_attention_large_seq_len_uniform_attention(self):
         device = torch.device("cuda")
         dtype = torch.bfloat16

torch/_inductor/compile_worker/subproc_pool.py

@@ -24,7 +24,6 @@ from typing_extensions import Never, ParamSpec
 import torch._thread_safe_fork  # noqa: F401
 from torch._inductor import config
 from torch._inductor.codecache import torch_key
-from torch._inductor.compile_worker.timer import Timer
 from torch._inductor.compile_worker.tracked_process_pool import (
     TrackedProcessPoolExecutor,
 )
@@ -133,7 +132,6 @@ class SubprocPool:
         nprocs: int,
         pickler: Optional[SubprocPickler] = None,
         kind: SubprocKind = SubprocKind.FORK,
-        quiesce: bool = False,
     ) -> None:
         entry = os.path.join(os.path.dirname(__file__), "__main__.py")
         self.pickler = pickler or SubprocPickler()
@@ -218,13 +216,6 @@ class SubprocPool:
             "pytorch.wait_counter.subproc_pool.first_job"
         ).guard()
 
-        if quiesce:
-            self.timer: Optional[Timer] = Timer(
-                config.quiesce_async_compile_time, self.quiesce
-            )
-        else:
-            self.timer = None
-
         # Start thread last to ensure all member variables are initialized
         # before any access.
         self.read_thread.start()
@@ -297,8 +288,6 @@ class SubprocPool:
             with self.futures_lock:
                 if not self.running:
                     return
-                if self.timer:
-                    self.timer.record_call()
                 if isinstance(result, _SubprocExceptionInfo):
                     # An exception occurred in the submitted job
                     self.pending_futures[job_id].set_exception(
@@ -333,8 +322,6 @@ class SubprocPool:
             with self.write_lock:
                 if not self.running:
                     return
-                if self.timer:
-                    self.timer.quit()
                 self.running = False
                 self.running_waitcounter.__exit__()
                 _send_msg(self.write_pipe, MsgHeader.SHUTDOWN)

torch/_inductor/compile_worker/timer.py

@@ -17,7 +17,7 @@ class Timer:
         self.background_thread: Optional[Thread] = None
         self.last_called: Optional[float] = None
         self.duration = duration
-        self.sleep_time = duration / 2
+        self.sleep_time = 60
         self.call = call
         self.exit = False
 

torch/_inductor/config.py

@@ -964,11 +964,6 @@ quiesce_async_compile_pool: bool = Config(
     default=False,
 )
 
-# Time in seconds to wait before quiescing
-quiesce_async_compile_time: int = Config(
-    default=60,
-)
-
 # Whether or not to enable statically launching CUDA kernels
 # compiled by triton (instead of using triton's own launcher)
 use_static_cuda_launcher: bool = static_cuda_launcher_default()