torchmonarch-nightly 2025.9.4__cp310-cp310-manylinux2014_x86_64.whl → 2025.9.5__cp310-cp310-manylinux2014_x86_64.whl

monarch/_src/actor/actor_mesh.py

@@ -863,6 +863,7 @@ class _Actor:
                     ins.rank = ctx.message_rank
                     try:
                         self.instance = Class(*args, **kwargs)
+                        self._maybe_exit_debugger()
                     except Exception as e:
                         self._saved_error = ActorError(
                             e, f"Remote actor {Class}.__init__ call failed."

monarch/_src/actor/host_mesh.py

@@ -35,11 +35,23 @@ def this_proc() -> "ProcMesh":
 
 
 def create_local_host_mesh() -> "HostMesh":
+    """
+    Create a local host mesh for the current machine.
+
+    Returns:
+        HostMesh: A single-host mesh configured for local process allocation.
+    """
     cmd, args, env = _get_bootstrap_args()
     return HostMesh(Shape.unity(), ProcessAllocator(cmd, args, env))
 
 
 class HostMesh(MeshTrait):
+    """
+    HostMesh represents a collection of compute hosts that can be used to spawn
+    processes and actors. The class requires you to provide your AllocateMixin that
+    interfaces with the underlying resource allocator of your choice.
+    """
+
     def __init__(self, shape: Shape, allocator: AllocateMixin):
         self._allocator = allocator
         self._shape = shape
@@ -57,12 +69,14 @@ class HostMesh(MeshTrait):
         """
         Start new processes on this host mesh. By default this starts one proc
         on each host in the mesh. Additional procs can be started using `per_host` to
-        specify the local shape, e.g.
+        specify the local shape, e.g.`
             per_host = {'gpus': 8}
         Will create a proc mesh with an additional 'gpus' dimension.
 
         `bootstrap` is a function that will be run at startup on each proc and can be used to e.g.
         configure CUDA or NCCL. We guarantee that CUDA has not been initialized before boostrap is called.
+
+        TODO: For now, a new allocator is created for every new ProcMesh.
         """
         if per_host is None:
             per_host = {}
@@ -107,6 +121,12 @@ class HostMesh(MeshTrait):
 
 
 def fake_in_process_host() -> "HostMesh":
+    """
+    Create a host mesh for testing and development using a local allocator.
+
+    Returns:
+        HostMesh: A host mesh configured with local allocation for in-process use.
+    """
     return HostMesh(Shape.unity(), LocalAllocator())
 
 

monarch/_src/actor/proc_mesh.py

@@ -95,14 +95,10 @@ if TYPE_CHECKING:
 
 class SetupActor(Actor):
     """
-    A helper actor to setup the proc mesh with user defined setup method.
-    Typically used to setup the environment variables.
+    A helper actor to set up the actor mesh with user defined setup method.
     """
 
     def __init__(self, env: Callable[[], None]) -> None:
-        """
-        Initialize the setup actor with the user defined setup method.
-        """
         self._setup_method = env
 
     @endpoint
@@ -133,8 +129,12 @@ def _use_standin_mesh() -> bool:
     return os.getenv("USE_STANDIN_ACTOR_MESH", default="0") != "0"
 
 
-# Ultra-hack to allow actors to identify proc meshes but with no real functionality.
 class ProcMeshRef:
+    """
+    A serializable remote reference to a ProcMesh. The reference is weak: No support
+    for refcount'ing. Spawning actors on a ProcMeshRef a stopped or a failed mesh will fail.
+    """
+
     def __init__(self, proc_mesh_id: int) -> None:
         self._proc_mesh_id = proc_mesh_id
         self._host_mesh: Optional["HostMesh"] = None
@@ -179,6 +179,17 @@ def _deref_proc_mesh(proc_mesh: ProcMeshRef) -> "ProcMesh":
 
 
 class ProcMesh(MeshTrait, DeprecatedNotAFuture):
+    """
+    A distributed mesh of processes for actor computation.
+
+    ProcMesh represents a collection of processes that can spawn and manage actors.
+    It provides the foundation for distributed actor systems by managing process
+    allocation, lifecycle, and communication across multiple hosts and devices.
+
+    The ProcMesh supports spawning actors, monitoring process health, logging
+    configuration, and code synchronization across distributed processes.
+    """
+
     def __init__(
         self,
         hy_proc_mesh: "Shared[HyProcMesh]",
@@ -249,6 +260,22 @@ class ProcMesh(MeshTrait, DeprecatedNotAFuture):
         return pm
 
     def spawn(self, name: str, Class: Type[T], *args: Any, **kwargs: Any) -> T:
+        """
+        Spawn a T-typed actor mesh on the process mesh.
+
+        Args:
+        - `name`: The name of the actor.
+        - `Class`: The class of the actor to spawn.
+        - `args`: Positional arguments to pass to the actor's constructor.
+        - `kwargs`: Keyword arguments to pass to the actor's constructor.
+
+        Returns:
+        - The actor instance.
+
+        Usage:
+            >>> procs: ProcMesh = host_mesh.spawn_procs(per_host={"gpus": 8})
+            >>> counters: Counter = procs.spawn("counters", Counter, 0)
+        """
         if self._slice:
             raise NotImplementedError("NYI: spawn on slice of a proc mesh.")
         return self._spawn_nonblocking(name, Class, *args, **kwargs)
@@ -294,19 +321,9 @@ class ProcMesh(MeshTrait, DeprecatedNotAFuture):
         Allocate a process mesh according to the provided alloc.
         Returns when the mesh is fully allocated.
 
-        Arguments:
-        - `alloc`: The alloc to allocate according to.
+        Args:
+        - `alloc`: A generator that yields a list of allocations.
         - `setup`: An optional lambda function to configure environment variables on the allocated mesh.
-        Use the `current_rank()` method within the lambda to obtain the rank.
-
-        Example of a setup method to initialize torch distributed environment variables:
-        ```
-        def setup():
-            rank = current_rank()
-            os.environ["RANK"] = str(rank)
-            os.environ["WORLD_SIZE"] = str(len(rank.shape))
-            os.environ["LOCAL_RANK"] = str(rank["gpus"])
-        ```
         """
 
         async def task() -> HyProcMesh:
@@ -432,6 +449,14 @@ class ProcMesh(MeshTrait, DeprecatedNotAFuture):
         conda: bool = False,
         auto_reload: bool = False,
     ) -> None:
+        """
+        Sync local code changes to the remote processes.
+
+        Args:
+            workspace: The workspace to sync.
+            conda: If True, also sync the currently activated conda env.
+            auto_reload: If True, automatically reload the workspace on changes.
+        """
         if self._code_sync_client is None:
             self._code_sync_client = CodeSyncMeshClient.spawn_blocking(
                 proc_mesh=await self._proc_mesh_for_asyncio_fixme,
@@ -525,6 +550,10 @@ class ProcMesh(MeshTrait, DeprecatedNotAFuture):
         return self
 
     def stop(self) -> Future[None]:
+        """
+        This will stop all processes (and actors) in the mesh and
+        release any resources associated with the mesh.
+        """
         self._logging_manager.stop()
 
         async def _stop_nonblocking() -> None:
@@ -574,6 +603,23 @@ class ProcMesh(MeshTrait, DeprecatedNotAFuture):
 
 
 def local_proc_mesh(*, gpus: Optional[int] = None, hosts: int = 1) -> ProcMesh:
+    """
+    Create a local process mesh for testing and development.
+
+    This function creates a process mesh using local allocation instead of
+    distributed process allocation. Primarily used for testing scenarios.
+
+    Args:
+        gpus: Number of GPUs to allocate per host. If None, uses local device count.
+        hosts: Number of hosts to allocate. Defaults to 1.
+
+    Returns:
+        ProcMesh: A locally allocated process mesh.
+
+    Warning:
+        This function is deprecated. Use `fake_in_process_host().spawn_procs()`
+        for testing or `this_proc().spawn_procs()` for current process actors.
+    """
     warnings.warn(
         "Use monarch._src.actor.host_mesh.fake_in_process_host().spawn_procs for testing. For launching an actor in the current process use this_proc().spawn_procs()",
         DeprecationWarning,
@@ -596,6 +642,22 @@ def sim_proc_mesh(
     dcs: int = 1,
     regions: int = 1,
 ) -> ProcMesh:
+    """Create a simulated process mesh for testing distributed scenarios.
+
+    This function creates a process mesh using simulation allocation to test
+    distributed behavior without requiring actual remote resources.
+
+    Args:
+        gpus: Number of GPUs per host. Defaults to 1.
+        hosts: Number of hosts. Defaults to 1.
+        racks: Number of racks. Defaults to 1.
+        zones: Number of zones. Defaults to 1.
+        dcs: Number of data centers. Defaults to 1.
+        regions: Number of regions. Defaults to 1.
+
+    Returns:
+        ProcMesh: A simulated process mesh with the specified topology.
+    """
     spec: AllocSpec = AllocSpec(
         AllocConstraints(),
         hosts=hosts,
@@ -658,6 +720,25 @@ def proc_mesh(
     env: dict[str, str] | None = None,
     setup: Callable[[], None] | None = None,
 ) -> ProcMesh:
+    """
+    Create a distributed process mesh across hosts.
+
+    This function creates a process mesh using distributed process allocation
+    across multiple hosts and GPUs. Used for production distributed computing.
+
+    Args:
+        gpus: Number of GPUs per host. If None, uses local device count.
+        hosts: Number of hosts to allocate. Defaults to 1.
+        env: Environment variables to set on remote processes.
+        setup: Optional setup function to run on each process at startup.
+
+    Returns:
+        ProcMesh: A distributed process mesh with the specified configuration.
+
+    Warning:
+        This function is deprecated. Use `this_host().spawn_procs()` with
+        appropriate per_host configuration instead.
+    """
     warnings.warn(
         "use this_host().spawn_procs(per_host = {'hosts': 2, 'gpus': 3}) instead of monarch.actor.proc_mesh(hosts=2, gpus=3)",
         DeprecationWarning,

monarch/actor/__init__.py

@@ -9,6 +9,7 @@
 Monarch Actor API - Public interface for actor functionality.
 """
 
+from monarch._rust_bindings.monarch_hyperactor.shape import Extent
 from monarch._src.actor.actor_mesh import (
     Accumulator,
     Actor,
@@ -19,6 +20,7 @@ from monarch._src.actor.actor_mesh import (
     current_actor_name,
     current_rank,
     current_size,
+    Endpoint,
     Point,
     Port,
     PortReceiver,
@@ -70,4 +72,6 @@ __all__ = [
     "hosts_from_config",
     "Port",
     "PortReceiver",
+    "Endpoint",
+    "Extent",
 ]

monarch/common/tensor.py

@@ -80,10 +80,6 @@ class Tensor(Referenceable, BaseTensor):
     in a device mesh. It provides the same interface as PyTorch tensors but
     enables distributed operations and communication patterns.
 
-    Args:
-        fake (torch.Tensor): A fake tensor representing the shape and type
-        mesh (DeviceMesh): The device mesh this tensor is distributed across
-        stream (Stream): The computation stream for this tensor
     """
 
     # pyre-fixme[13]: Attribute `stream` is never initialized.

{torchmonarch_nightly-2025.9.4.dist-info → torchmonarch_nightly-2025.9.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: torchmonarch-nightly
-Version: 2025.9.4
+Version: 2025.9.5
 Summary: Monarch: Single controller library
 Author: Meta
 Author-email: oncall+monarch@xmail.facebook.com
@@ -9,6 +9,7 @@ Requires-Python: >= 3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: torch
+Requires-Dist: torchshow
 Requires-Dist: pyzmq
 Requires-Dist: requests
 Requires-Dist: numpy
@@ -73,18 +74,24 @@ The [introduction to monarch concepts](https://meta-pytorch.org/monarch/generate
 > work. It's recommended that you signal your intention to contribute in the
 > issue tracker, either by filing a new issue or by claiming an existing one.
 
-Note: Monarch is currently only supported on Linux systems
-
 ## 📖 Documentation
 
 View Monarch's hosted documentation [at this link](https://meta-pytorch.org/monarch/).
 
 ## Installation
+Note for running distributed tensors, the local torch version must match the version that monarch was built with.
 
 ### On Fedora distributions
+## Stable
+`pip install torchmonarch`
+
+torchmonarch stable is built with the latest stable torch.
 
+## Nightly
 `pip install torchmonarch-nightly`
 
+torchmonarch-nightly is built with torch nightly.
+
 or manually
 
 ```sh

{torchmonarch_nightly-2025.9.4.dist-info → torchmonarch_nightly-2025.9.5.dist-info}/RECORD

@@ -1,5 +1,5 @@
 monarch/__init__.py,sha256=mgKiyD1kxky-1pvhMlNfF4VmxWnhi-FSYZNFzkW1BEM,7052
-monarch/_rust_bindings.so,sha256=TlxG4eDuLNUGhFc98lgTTF5IgAjUBFbm6MDmue-HvmY,61368760
+monarch/_rust_bindings.so,sha256=BD_ufqVQWW4VoQcXui0hydKgOlh8xsLSEJzG-W_UJIo,61366688
 monarch/_testing.py,sha256=5BDMVA4hBMo780rsJ39vRmUZi6mTN8aYY7I9grJRjJ8,7841
 monarch/actor_mesh.py,sha256=VtPU9syi_vUdwDSJJ639Z4Y_EcWZUScyoj0lQ88RQPs,421
 monarch/bootstrap_main.py,sha256=39OZpNMrfvvNJf-iwuNzgslzYA_ItaRPHfXGn_V74N0,524
@@ -8,7 +8,7 @@ monarch/fetch.py,sha256=CssP25dMqyJnJAWoC41lwkMnSbvS-f2DL9PRbudJXfc,1704
 monarch/gradient_generator.py,sha256=b7PmoN_F3c5hQglfHeW_v5htYnePKvJGkzZN-tpHR4A,6396
 monarch/memory.py,sha256=ol86dBhFAJqg78iF25-BuK0wuwj1onR8FIioZ_B0gjw,1377
 monarch/mesh_controller.py,sha256=Y_26Cnmp72TccNbWdDQhq18j7de7pSw83E_fREJX9Yo,15372
-monarch/monarch_controller,sha256=Ee-Lv-fM19tmWazjxrIGcpH6UPuom701xIMhuykaflA,32385432
+monarch/monarch_controller,sha256=ag3n7pLBf1bfo5fX09nohswRiOsrLWQsJ4ZqvVSsh7Y,32442720
 monarch/notebook.py,sha256=zu9MKDFKf1-rCM2TqFSRJjMBeiWuKcJSyUFLvoZRQzs,25949
 monarch/opaque_module.py,sha256=jCcg0DjbcEVXA9WNG0NhUzGteLHOJLTZEBvrIYJIAns,10436
 monarch/opaque_object.py,sha256=x1LoX6RIMGh4ux52xIfhPgoh6PhZHdkf9bMccHW3DW0,2808
@@ -25,17 +25,17 @@ monarch/tensorboard.py,sha256=MnLgH5lbqeUJauEuirEgR6L_qYl2NGdtwZOWIAuOZao,2587
 monarch/world_mesh.py,sha256=ob5dJWaC49Uw0xqClHBm8CQLvL4xKnjd4TGzk7k8NxI,980
 monarch/_src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 monarch/_src/actor/__init__.py,sha256=4iK3nzQZmEPe0HGNY70fABBenw3lCVVaaF0xddF5Fa0,235
-monarch/_src/actor/actor_mesh.py,sha256=QH1ITgg9pCRg6LAmdFKhPAYiuUKN757vZUslXscUbNM,40474
+monarch/_src/actor/actor_mesh.py,sha256=P9b4CvxYeYfJecPxeEtohAasvGgzzElcvxh9chALeAA,40526
 monarch/_src/actor/allocator.py,sha256=UVGhrkPQMqPQp6vUngPI361s6yCEfZ0gfz8WTtG2om4,9392
 monarch/_src/actor/bootstrap_main.py,sha256=7T7ARumcHLZ5RI-k5jej9tBil0J7-BUSVFKwAZO2tJU,2413
 monarch/_src/actor/device_utils.py,sha256=gBpl23wMjppVAEzzj8U9HyX-B7Bs2_3ftiMAkzUS4j4,577
 monarch/_src/actor/endpoint.py,sha256=_VaPHc0Fcj1P5nDzUXt8xnS6iw7-HO6hGx7W5RtU3eU,10916
 monarch/_src/actor/event_loop.py,sha256=2i4fKIkemBzua_t47BqVa2roZ6fWB6sbmMFPNx2zKN0,2832
 monarch/_src/actor/future.py,sha256=idgqzU_B5qWfClIP5dTLapmBflWq5va-ujAzUbT1Asc,7490
-monarch/_src/actor/host_mesh.py,sha256=xSto21T4ESnF55UfcQOkWZxsIiiv3VnOzyrcrZ_or7Y,4405
+monarch/_src/actor/host_mesh.py,sha256=8SOkg_LhHuzLyhpwxT7Yw1_h8QrIlwfWhrSwHyAvfnk,5083
 monarch/_src/actor/logging.py,sha256=9aguohqCtvLVwWGTFM7o-rBptT26BjI2s6E5Of2imM4,3311
 monarch/_src/actor/pickle.py,sha256=FhdbAEsGrsD7f25bxF7HlROLm6j2TTvmToq8P1kyhB8,2913
-monarch/_src/actor/proc_mesh.py,sha256=fijaxhvVn1adtPFPaZ_-16pvWo1-rhjbB6qHkW2dAkk,24779
+monarch/_src/actor/proc_mesh.py,sha256=lYrRMQNOGAdFXuFvc3lQ68xIS01YJWMkpi8qH5HHAHE,27791
 monarch/_src/actor/python_extension_methods.py,sha256=QujLWOQQbDdGCin8tZfDxyIwkM-Md4t9QtcTGTHOE_s,3493
 monarch/_src/actor/shape.py,sha256=PJqxpQEISHlxK8rrlKWpcNMEHiGxBbc6TsHcGZCOsyE,8472
 monarch/_src/actor/source_loader.py,sha256=TGHmExLyxPDcCyuG254zo6aUqHMpl-j0VWzxa9rkJYQ,1405
@@ -52,7 +52,7 @@ monarch/_src/debug_cli/__init__.py,sha256=NNrKh5KdiYdbxOhin8x-gw_-tvcuGex2UbS_z7
 monarch/_src/debug_cli/debug_cli.py,sha256=OJqqVFXcMkj-bnrxcE2VnjIgA5xrlKjEtCstrsdPcm0,1146
 monarch/_src/tensor_engine/__init__.py,sha256=Md3cCHD7Ano9kV15PqGbicgUO-RMdh4aVy1yKiDt_xE,208
 monarch/_src/tensor_engine/rdma.py,sha256=62saqcXo6oUxH9rEZShNGLYdRZ_zizLQKhub7LDSaBg,8418
-monarch/actor/__init__.py,sha256=9CXerFsBVR-lcpSEw_lIsNNP0qqCWoSA9exyDhhGvHI,1458
+monarch/actor/__init__.py,sha256=hHf8ri2czQwi-Z23Z9cYZ2FvkVbYOcDA_GTLW_rju7k,1569
 monarch/builtins/__init__.py,sha256=QcfnHZGbc2qktBg7DyZt2ruE6VahnIt4S8lEZLHdJqU,443
 monarch/builtins/log.py,sha256=H1QkuVzwxyi36Zyv-XR0VN0QsNimBWwxE1__fjs0_2o,554
 monarch/builtins/random.py,sha256=wPbvscg7u53EXpMFo885fO2XOlsyjrNAJ4rBxLzfxdg,1839
@@ -84,7 +84,7 @@ monarch/common/reference.py,sha256=O26lkzEeVwj0S1xEy-OLqdHVnACmmlbQCUmXRrW4n1Q,9
 monarch/common/remote.py,sha256=uc2JUbYHMnEZRnw9ZuS4mgvK_UHCuRaIIFbzZpx89hQ,11485
 monarch/common/selection.py,sha256=lpWFbZs3ArYy29e-53eoAVAjQFksf1RvZz9NvM0CUW4,308
 monarch/common/stream.py,sha256=_ejoxafHtdD10lLzznRCXKwrkZ_ZH9k_VTgiA5yfBrI,3583
-monarch/common/tensor.py,sha256=tqoXyvxrmfTtOm6ToWoy9NVg1NCjlt1KPjrNtYOMCco,29824
+monarch/common/tensor.py,sha256=9FIUn5--VeacYWhEhMvysdG0yc_zq4eW3X3526RCw3w,29598
 monarch/common/tensor_factory.py,sha256=qm8NZx-5ezMAFjNLiXQvb66okm5XgdboB_GRarGOdN0,801
 monarch/common/tree.py,sha256=1DG3siiE7ixBV6v5cwN8RT_17aJhYZTE-L3i7wZe2_c,2282
 monarch/controller/__init__.py,sha256=J8qjUOysmcMAek2KFN13mViOXZxTYc5vCrF02t3VuFU,223
@@ -98,7 +98,7 @@ monarch/debug_cli/__init__.py,sha256=NNrKh5KdiYdbxOhin8x-gw_-tvcuGex2UbS_z7MV9g0
 monarch/debug_cli/__main__.py,sha256=FGsQn54RkC_3gpRrm_UFrGiDDHRbMeGzXXsGANr5UHU,317
 monarch/gradient/__init__.py,sha256=kqmzwt16mMpk0M3GhpgP_f7da4DGnaV9chDzbt66k4Q,308
 monarch/gradient/_gradient_generator.pyi,sha256=6cX0UxaDt9NAlwgIhTgnweqGOf6qRhHiGnUzSWNCxdU,630
-monarch/gradient/_gradient_generator.so,sha256=VjKG-8JGdoIado7GM8ohLiBxxD99GXCO1x2lEdKSLGU,12106080
+monarch/gradient/_gradient_generator.so,sha256=RQ6LyhpUPyq3qKTr3TTJftbQC0pVALk0z_fd6pAlrMc,12108480
 monarch/parallel/__init__.py,sha256=6920kIkhiX7AiyjYvyc1ad8ccP-bStJJ1sS5KkeN2P0,352
 monarch/parallel/pipelining/__init__.py,sha256=J8qjUOysmcMAek2KFN13mViOXZxTYc5vCrF02t3VuFU,223
 monarch/parallel/pipelining/runtime.py,sha256=KK8TG1gUYEzSsquiZoPTWGSIC74mlncD7cYknKxfb3c,32470
@@ -186,9 +186,9 @@ tests/simulator/test_profiling.py,sha256=TGYCfzTLdkpIwnOuO6KApprmrgPIRQe60KRX3wk
 tests/simulator/test_simulator.py,sha256=LO8lA0ssY-OGEBL5ipEu74f97Y765TEwfUOv-DtIptM,14568
 tests/simulator/test_task.py,sha256=ipqBDuDAysuo1xOB9S5psaFvwe6VATD43IovCTSs0t4,2327
 tests/simulator/test_worker.py,sha256=QrWWIJ3HDgDLkBPRc2mwYPlOQoXQcj1qRfc0WUfKkFY,3507
-torchmonarch_nightly-2025.9.4.dist-info/licenses/LICENSE,sha256=e0Eotbf_rHOYPuEUlppIbvwy4SN98CZnl_hqwvbDA4Q,1530
-torchmonarch_nightly-2025.9.4.dist-info/METADATA,sha256=mRWehYrs_QENvKeR77pdQ8AO9LIPeGwd1Tarqqf05gs,6236
-torchmonarch_nightly-2025.9.4.dist-info/WHEEL,sha256=_wZSFk0d90K9wOBp8Q-UGxshyiJ987JoPiyUBNC6VLk,104
-torchmonarch_nightly-2025.9.4.dist-info/entry_points.txt,sha256=60QVSpYVzkzS4iDOiLp0fsLxVp47X3J2l3v7W-59LMo,117
-torchmonarch_nightly-2025.9.4.dist-info/top_level.txt,sha256=E-ZssZzyM17glpVrh-S9--qJ-w9p2EjuYOuNw9tQ4Eg,33
-torchmonarch_nightly-2025.9.4.dist-info/RECORD,,
+torchmonarch_nightly-2025.9.5.dist-info/licenses/LICENSE,sha256=e0Eotbf_rHOYPuEUlppIbvwy4SN98CZnl_hqwvbDA4Q,1530
+torchmonarch_nightly-2025.9.5.dist-info/METADATA,sha256=6WKzpl0pCJnxLhSxTckSsA7C5ncuUZJ2_NzujIuxhiQ,6474
+torchmonarch_nightly-2025.9.5.dist-info/WHEEL,sha256=_wZSFk0d90K9wOBp8Q-UGxshyiJ987JoPiyUBNC6VLk,104
+torchmonarch_nightly-2025.9.5.dist-info/entry_points.txt,sha256=60QVSpYVzkzS4iDOiLp0fsLxVp47X3J2l3v7W-59LMo,117
+torchmonarch_nightly-2025.9.5.dist-info/top_level.txt,sha256=E-ZssZzyM17glpVrh-S9--qJ-w9p2EjuYOuNw9tQ4Eg,33
+torchmonarch_nightly-2025.9.5.dist-info/RECORD,,