torchx-nightly 2024.1.29__py3-none-any.whl → 2024.1.31__py3-none-any.whl

torchx/components/dist.py

@@ -174,6 +174,7 @@ def ddp(
     rdzv_backend: str = "c10d",
     mounts: Optional[List[str]] = None,
     debug: bool = False,
+    tee: int = 3,
 ) -> specs.AppDef:
     """
     Distributed data parallel style application (one role, multi-replica).
@@ -208,6 +209,7 @@ def ddp(
         mounts: mounts to mount into the worker environment/container (ex. type=<bind/volume>,src=/host,dst=/job[,readonly]).
                 See scheduler documentation for more info.
         debug: whether to run with preset debug flags enabled
+        tee: tees the specified std stream(s) to console + file. 0: none, 1: stdout, 2: stderr, 3: both
     """
 
     if (script is None) == (m is None):
@@ -262,7 +264,7 @@ def ddp(
         "--nproc_per_node",
         str(nproc_per_node),
         "--tee",
-        "3",
+        str(tee),
         "--role",
         "",
     ]

torchx/schedulers/gcp_batch_scheduler.py

@@ -205,14 +205,12 @@ class GCPBatchScheduler(Scheduler[GCPBatchOpts]):
             if cpu <= 0:
                 cpu = 1
             MILLI = 1000
-            # pyre-fixme[8]: Attribute has type `Field`; used as `int`.
             res.cpu_milli = cpu * MILLI
             memMB = resource.memMB
             if memMB < 0:
                 raise ValueError(
                     f"memMB should to be set to a positive value, got {memMB}"
                 )
-            # pyre-fixme[8]: Attribute has type `Field`; used as `int`.
             res.memory_mib = memMB
 
             # TODO support named resources
@@ -360,13 +358,11 @@ class GCPBatchScheduler(Scheduler[GCPBatchOpts]):
             return None
 
         gpu = 0
-        # pyre-fixme[16]: `Field` has no attribute `instances`.
         if len(job.allocation_policy.instances) != 0:
             gpu_type = job.allocation_policy.instances[0].policy.machine_type
             gpu = GPU_TYPE_TO_COUNT[gpu_type]
 
         roles = {}
-        # pyre-fixme[16]: `RepeatedField` has no attribute `__iter__`.
         for tg in job.task_groups:
             env = tg.task_spec.environment.variables
             role = env["TORCHX_ROLE_NAME"]
@@ -390,7 +386,6 @@ class GCPBatchScheduler(Scheduler[GCPBatchOpts]):
         # TODO map role/replica status
         desc = DescribeAppResponse(
             app_id=app_id,
-            # pyre-fixme[16]: `Field` has no attribute `state`.
             state=JOB_STATE[job.status.state.name],
             roles=list(roles.values()),
         )

torchx/workspace/docker_workspace.py

@@ -105,6 +105,10 @@ class DockerWorkspaceMixin(WorkspaceMixin[Dict[str, Tuple[str, str]]]):
             workspace: a fsspec path to a directory with contents to be overlaid
         """
 
+        old_imgs = [
+            image.id
+            for image in self._docker_client.images.list(name=cfg["image_repo"])
+        ]
         context = _build_context(role.image, workspace)
 
         try:
@@ -129,7 +133,8 @@ class DockerWorkspaceMixin(WorkspaceMixin[Dict[str, Tuple[str, str]]]):
                     self.LABEL_VERSION: torchx.__version__,
                 },
             )
-            role.image = image.id
+            if len(old_imgs) == 0 or role.image not in old_imgs:
+                role.image = image.id
         finally:
             context.close()
 

{torchx_nightly-2024.1.29.dist-info → torchx_nightly-2024.1.31.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: torchx-nightly
-Version: 2024.1.29
+Version: 2024.1.31
 Summary: TorchX SDK and Components
 Home-page: https://github.com/pytorch/torchx
 Author: TorchX Devs

{torchx_nightly-2024.1.29.dist-info → torchx_nightly-2024.1.31.dist-info}/RECORD

@@ -24,7 +24,7 @@ torchx/cli/colors.py,sha256=bVN_jEDwLgvypnDMeCHKn0q0ZDDhQjBJnyVfZHAE6nc,553
 torchx/cli/main.py,sha256=DJYikTWacADa4VoscqZGjZmMKWWK29tBl6-pGtnzsRE,3469
 torchx/components/__init__.py,sha256=6-TQ4SY-Tn56os_1lOs_HMabOoE7gkkud_8e1BgvfJw,12106
 torchx/components/component_test_base.py,sha256=eKOwBp5cRgiA4FgZd_FCvyJ-ppv2v3JN9AGXnaSK_Cw,4135
-torchx/components/dist.py,sha256=-YqNOx1XdcDhleqUJI0xgCOxZWj-h5yISBFIMPd-wm8,14428
+torchx/components/dist.py,sha256=tBOL_DjUeBjNDUcF2wnjlUEzZck9fo1ojRkbQA5ERT8,14555
 torchx/components/interpret.py,sha256=g8gkKdDJvsBfX1ZrpVT7n2bMEtmwRV_1AqDyAnnQ_aA,697
 torchx/components/metrics.py,sha256=1gbp8BfzZWGa7PD1db5vRADlONzmae4qSBUUdCWayr0,2814
 torchx/components/serve.py,sha256=9RlpwlU2KOC7sMOZBeYwUpJIKDCXrU8xNo1SH-AT3fc,2141
@@ -68,7 +68,7 @@ torchx/schedulers/api.py,sha256=XlYrD6ZjV71HotJxdVZxA_Zc8DuxhM4KKCnkibqZflU,1414
 torchx/schedulers/aws_batch_scheduler.py,sha256=t6wGK1NUjhTAoxH3ie7lIMTl2cxwdixkcT3HxKsggDk,27517
 torchx/schedulers/devices.py,sha256=PNbcpf8fEM18Ag1RgK9Q30zPBalEcPdsFWctdbLxuv8,1352
 torchx/schedulers/docker_scheduler.py,sha256=I-kZN-dXoQyokLPe9ZKjfhkVX5lHx_C5jvLLc2JmXQQ,15456
-torchx/schedulers/gcp_batch_scheduler.py,sha256=mBxJbrNTUbIuYmudzyhOOcf8KAuUpxhiQMTDgJPtL8M,16549
+torchx/schedulers/gcp_batch_scheduler.py,sha256=pgPnlx5tzogPw6FuOkoaxBY-rVjEh3mWFP-yX5izGZ8,16206
 torchx/schedulers/ids.py,sha256=IGsJEbCYTdfKdU3MhKLQU6b7sWCJy5dlRV6JIL_9BlE,1783
 torchx/schedulers/kubernetes_mcad_scheduler.py,sha256=xAt-on3K8HwS2kzWasn0zXd2q4IDQzo2N5A5Ehh9NII,42885
 torchx/schedulers/kubernetes_scheduler.py,sha256=6NXYJwiYCXNeB3ubr8t4q_SuAa-vlYdiCAPXTB3f-zg,27068
@@ -110,10 +110,10 @@ torchx/util/types.py,sha256=6ASuDKGO91UU3DCSuWhPX_C03341tApLCQEByUz8xpY,7016
 torchx/workspace/__init__.py,sha256=KbGEzJqqXaIxALm_EQO64aw-fE7MeDMFXcpU1mY650I,783
 torchx/workspace/api.py,sha256=Ej6DR__mNWaVyZgoVNAAOloDy1kTD5X1jz7pRtoVf80,5464
 torchx/workspace/dir_workspace.py,sha256=Fz-hKIx0KN8iJf2BsthNj0NvTkWlxP6WFsElPs_BaT0,2253
-torchx/workspace/docker_workspace.py,sha256=Yd8ut26bNfjyJQnmH8ANOrflfr-4VKcnOrIjbi_XIUY,9208
-torchx_nightly-2024.1.29.dist-info/LICENSE,sha256=WVHfXhFC0Ia8LTKt_nJVYobdqTJVg_4J3Crrfm2A8KQ,1721
-torchx_nightly-2024.1.29.dist-info/METADATA,sha256=SCFwHZDtkGq7XCsEuQlfXljWwmCpl5X1s01qLXm_abk,5611
-torchx_nightly-2024.1.29.dist-info/WHEEL,sha256=oiQVh_5PnQM0E3gPdiz09WCNmwiHDMaGer_elqB3coM,92
-torchx_nightly-2024.1.29.dist-info/entry_points.txt,sha256=3JYZFlX9aWzR-Gs_qsx1zq7mlqbFz6Mi9rQUULW8caI,170
-torchx_nightly-2024.1.29.dist-info/top_level.txt,sha256=pxew3bc2gsiViS0zADs0jb6kC5v8o_Yy_85fhHj_J1A,7
-torchx_nightly-2024.1.29.dist-info/RECORD,,
+torchx/workspace/docker_workspace.py,sha256=cqvxHTtrVza0gkoImbzeRiroWkCjdBXRVgONvaulS0g,9410
+torchx_nightly-2024.1.31.dist-info/LICENSE,sha256=WVHfXhFC0Ia8LTKt_nJVYobdqTJVg_4J3Crrfm2A8KQ,1721
+torchx_nightly-2024.1.31.dist-info/METADATA,sha256=07GcG1r70hTFIjIOPpGN88EQFp3oQ7TIRwFvEl4lSkI,5611
+torchx_nightly-2024.1.31.dist-info/WHEEL,sha256=oiQVh_5PnQM0E3gPdiz09WCNmwiHDMaGer_elqB3coM,92
+torchx_nightly-2024.1.31.dist-info/entry_points.txt,sha256=3JYZFlX9aWzR-Gs_qsx1zq7mlqbFz6Mi9rQUULW8caI,170
+torchx_nightly-2024.1.31.dist-info/top_level.txt,sha256=pxew3bc2gsiViS0zADs0jb6kC5v8o_Yy_85fhHj_J1A,7
+torchx_nightly-2024.1.31.dist-info/RECORD,,