wandb 0.19.12rc1__py3-none-win32.whl → 0.20.1__py3-none-win32.whl

wandb/sdk/wandb_settings.py

@@ -3,7 +3,6 @@ from __future__ import annotations
 import configparser
 import json
 import logging
-import multiprocessing
 import os
 import pathlib
 import platform
@@ -25,7 +24,7 @@ from pydantic import BaseModel, ConfigDict, Field
 from typing_extensions import Self
 
 import wandb
-from wandb import env, termwarn, util
+from wandb import env, util
 from wandb._pydantic import (
     IS_PYDANTIC_V2,
     AliasChoices,
@@ -529,11 +528,6 @@ class Settings(BaseModel, validate_assignment=True):
     x_disable_meta: bool = False
     """Flag to disable the collection of system metadata."""
 
-    x_disable_service: bool = False
-    """Flag to disable the W&B service.
-
-    This is deprecated and will be removed in future versions."""
-
     x_disable_setproctitle: bool = False
     """Flag to disable using setproctitle for the internal process in the legacy service.
 
@@ -704,6 +698,15 @@ class Settings(BaseModel, validate_assignment=True):
     x_service_wait: float = 30.0
     """Time in seconds to wait for the wandb-core internal service to start."""
 
+    x_skip_transaction_log: bool = False
+    """Whether to skip saving the run events to the transaction log.
+
+    This is only relevant for online runs. Can be used to reduce the amount of
+    data written to disk.
+
+    Should be used with caution, as it removes the gurantees about
+    recoverability."""
+
     x_start_time: Optional[float] = None
     """The start time of the run in seconds since the Unix epoch."""
 
@@ -769,6 +772,19 @@ class Settings(BaseModel, validate_assignment=True):
     Can be accessed via run._system_metrics.
     """
 
+    x_stats_coreweave_metadata_base_url: str = "http://169.254.169.254"
+    """The scheme and hostname for contacting the CoreWeave metadata server.
+
+    Only accessible from within a CoreWeave cluster.
+    """
+
+    x_stats_coreweave_metadata_endpoint: str = "/api/v2/cloud-init/meta-data"
+    """The relative path on the CoreWeave metadata server to which to make requests.
+
+    This must not include the schema and hostname prefix.
+    Only accessible from within a CoreWeave cluster.
+    """
+
     x_sync: bool = False
     """Flag to indicate whether we are syncing a run from the transaction log."""
 
@@ -812,6 +828,12 @@ class Settings(BaseModel, validate_assignment=True):
                     "Please specify only one of them."
                 )
             return self
+
+        @model_validator(mode="after")
+        def validate_skip_transaction_log(self):
+            if self._offline and self.x_skip_transaction_log:
+                raise ValueError("Cannot skip transaction log in offline mode")
+            return self
     else:
 
         @root_validator(pre=False)  # type: ignore [call-overload]
@@ -830,19 +852,14 @@ class Settings(BaseModel, validate_assignment=True):
                 )
             return values
 
-    # Field validators.
-
-    @field_validator("x_disable_service", mode="after")
-    @classmethod
-    def validate_disable_service(cls, value):
-        if value:
-            termwarn(
-                "Disabling the wandb service is deprecated as of version 0.18.0 "
-                "and will be removed in future versions. ",
-                repeat=False,
-            )
-        return value
+        @root_validator(pre=False)  # type: ignore [call-overload]
+        @classmethod
+        def validate_skip_transaction_log(cls, values):
+            if values.get("_offline") and values.get("x_skip_transaction_log"):
+                raise ValueError("Cannot skip transaction log in offline mode")
+            return values
 
+    # Field validators.
     @field_validator("api_key", mode="after")
     @classmethod
     def validate_api_key(cls, value):
@@ -878,23 +895,7 @@ class Settings(BaseModel, validate_assignment=True):
         if value != "auto":
             return value
 
-        if hasattr(values, "data"):
-            # pydantic v2
-            values = values.data
-        else:
-            # pydantic v1
-            values = values
-
-        if (
-            ipython.in_jupyter()
-            or (values.get("start_method") == "thread")
-            or not values.get("x_disable_service")
-            or platform.system() == "Windows"
-        ):
-            value = "wrap"
-        else:
-            value = "redirect"
-        return value
+        return "wrap"
 
     @field_validator("x_executable", mode="before")
     @classmethod
@@ -1080,15 +1081,19 @@ class Settings(BaseModel, validate_assignment=True):
     def validate_start_method(cls, value):
         if value is None:
             return value
-        available_methods = ["thread"]
-        if hasattr(multiprocessing, "get_all_start_methods"):
-            available_methods += multiprocessing.get_all_start_methods()
-        if value not in available_methods:
-            raise UsageError(
-                f"Settings field `start_method`: {value!r} not in {available_methods}"
-            )
+        wandb.termwarn(
+            "`start_method` is deprecated and will be removed in a future version "
+            "of wandb. This setting is currently non-functional and safely ignored.",
+            repeat=False,
+        )
         return value
 
+    @field_validator("x_stats_coreweave_metadata_base_url", mode="after")
+    @classmethod
+    def validate_x_stats_coreweave_metadata_base_url(cls, value):
+        validate_url(value)
+        return value.rstrip("/")
+
     @field_validator("x_stats_gpu_device_ids", mode="before")
     @classmethod
     def validate_x_stats_gpu_device_ids(cls, value):
@@ -1441,7 +1446,6 @@ class Settings(BaseModel, validate_assignment=True):
         env_prefix: str = "WANDB_"
         private_env_prefix: str = env_prefix + "_"
         special_env_var_names = {
-            "WANDB_DISABLE_SERVICE": "x_disable_service",
             "WANDB_SERVICE_TRANSPORT": "x_service_transport",
             "WANDB_DIR": "root_dir",
             "WANDB_NAME": "run_name",
@@ -1637,21 +1641,7 @@ class Settings(BaseModel, validate_assignment=True):
 
     def _get_program(self) -> Optional[str]:
         """Get the program that started the current process."""
-        if not self._jupyter:
-            # If not in a notebook, try to get the program from the environment
-            # or the __main__ module for scripts run as `python -m ...`.
-            program = os.getenv(env.PROGRAM)
-            if program is not None:
-                return program
-            try:
-                import __main__
-
-                if __main__.__spec__ is None:
-                    return __main__.__file__
-                return f"-m {__main__.__spec__.name}"
-            except (ImportError, AttributeError):
-                return None
-        else:
+        if self._jupyter:
             # If in a notebook, try to get the program from the notebook metadata.
             if self.notebook_name:
                 return self.notebook_name
@@ -1661,8 +1651,29 @@ class Settings(BaseModel, validate_assignment=True):
 
             if self.x_jupyter_path.startswith("fileId="):
                 return self.x_jupyter_name
+
+            return self.x_jupyter_path
+
+        # If not in a notebook, try to get the program from the environment
+        # or the __main__ module for scripts run as `python -m ...`.
+        program = os.getenv(env.PROGRAM)
+        if program is not None:
+            return program
+
+        try:
+            import __main__
+        except ImportError:
+            return None
+
+        try:
+            if __main__.__spec__ is None:
+                python_args = __main__.__file__
             else:
-                return self.x_jupyter_path
+                python_args = f"-m {__main__.__spec__.name}"
+        except AttributeError:
+            return None
+
+        return python_args
 
     @staticmethod
     def _get_program_relpath(program: str, root: Optional[str] = None) -> Optional[str]:

wandb/sdk/wandb_setup.py

@@ -1,14 +1,25 @@
-#
-"""Setup wandb session.
-
-This module configures a wandb session which can extend to multiple wandb runs.
-
-Functions:
-    setup(): Configure wandb session.
-
-Early logging keeps track of logger output until the call to wandb.init() when the
-run_id can be resolved.
-
+"""Global W&B library state.
+
+This module manages global state, which for wandb includes:
+
+- Settings configured through `wandb.setup()`
+- The list of active runs
+- A subprocess ("the internal service") that asynchronously uploads metrics
+
+This module is fork-aware: in a forked process such as that spawned by the
+`multiprocessing` module, `wandb.singleton()` returns a new object, not the
+one inherited from the parent process. This requirement comes from backward
+compatibility with old design choices: the hardest one to fix is that wandb
+was originally designed to have a single run for the entire process that
+`wandb.init()` was meant to return. Back then, the only way to create
+multiple simultaneous runs in a single script was to run subprocesses, and since
+the built-in `multiprocessing` module forks by default, this required a PID
+check to make `wandb.init()` ignore the inherited global run.
+
+Another reason for fork-awareness is that the process that starts up
+the internal service owns it and is responsible for shutting it down,
+and child processes shouldn't also try to do that. This is easier to
+redesign.
 """
 
 from __future__ import annotations
@@ -344,10 +355,25 @@ The value is invalid and must not be used if `os.getpid() != _singleton._pid`.
 """
 
 
-def singleton() -> _WandbSetup | None:
-    """Returns the W&B singleton if it exists for the current process.
+def singleton() -> _WandbSetup:
+    """The W&B singleton for the current process.
+
+    The first call to this in this process (which may be a fork of another
+    process) creates the singleton, and all subsequent calls return it
+    until teardown(). This does not start the service process.
+    """
+    return _setup(start_service=False)
+
+
+def singleton_if_setup() -> _WandbSetup | None:
+    """The W&B singleton for the current process or None if it isn't set up.
+
+    Always prefer singleton() over this function.
 
-    Unlike setup(), this does not create the singleton if it doesn't exist.
+    Unlike singleton(), this never creates the singleton and therefore never
+    initializes global settings from the environment. This is useful only
+    during tests, which may modify the environment after having imported wandb
+    and called certain functions.
     """
     if _singleton and _singleton._pid == os.getpid():
         return _singleton

wandb/sdk/wandb_watch.py

@@ -59,9 +59,8 @@ def _watch(
             The graph object, which will be populated after the first backward pass.
 
     Raises:
-        ValueError:
-            If `wandb.init` has not been called or if any of the models are not instances
-            of `torch.nn.Module`.
+        ValueError: If `wandb.init` has not been called.
+        TypeError: If any of the models are not instances of `torch.nn.Module`.
     """
     global _global_watch_idx
 
@@ -85,9 +84,8 @@ def _watch(
 
     for model in models:
         if not isinstance(model, torch.nn.Module):
-            raise ValueError(
-                "Expected a pytorch model (torch.nn.Module). Received "
-                + str(type(model))
+            raise TypeError(
+                f"Expected a pytorch model (torch.nn.Module). Received {type(model)}"
             )
 
     graphs = []
@@ -139,7 +137,7 @@ def _unwatch(
             models = (models,)
         for model in models:
             if not hasattr(model, "_wandb_hook_names"):
-                wandb.termwarn("{} model has not been watched".format(model))
+                wandb.termwarn(f"{model} model has not been watched")
             else:
                 for name in model._wandb_hook_names:
                     run._torch.unhook(name)

wandb/sync/__init__.py

@@ -1,3 +1,3 @@
 """module sync."""
 
-from .sync import SyncManager, get_run_from_path, get_runs  # noqa: F401
+from .sync import SyncManager, get_run_from_path, get_runs

wandb/sync/sync.py

@@ -157,13 +157,13 @@ class SyncThread(threading.Thread):
         proto_run.entity = self._entity
         proto_run.telemetry.feature.sync_tfevents = True
 
-        url = "{}/{}/{}/runs/{}".format(
-            self._app_url,
-            url_quote(proto_run.entity),
-            url_quote(proto_run.project),
-            url_quote(proto_run.run_id),
+        url = (
+            f"{self._app_url}"
+            f"/{url_quote(proto_run.entity)}"
+            f"/{url_quote(proto_run.project)}"
+            f"/runs/{url_quote(proto_run.run_id)}"
         )
-        print("Syncing: {} ...".format(url))  # noqa: T201
+        print(f"Syncing: {url} ...")  # noqa: T201
         sys.stdout.flush()
         # using a handler here automatically handles the step
         # logic, adds summaries to the run, and handles different
@@ -241,7 +241,7 @@ class SyncThread(threading.Thread):
                 )
                 return None
             else:
-                raise e
+                raise
 
     def run(self):
         if self._log_path is not None:
@@ -305,13 +305,13 @@ class SyncThread(threading.Thread):
                     if not shown and result_type == "run_result":
                         r = result.run_result.run
                         # TODO(jhr): hardcode until we have settings in sync
-                        url = "{}/{}/{}/runs/{}".format(
-                            self._app_url,
-                            url_quote(r.entity),
-                            url_quote(r.project),
-                            url_quote(r.run_id),
+                        url = (
+                            f"{self._app_url}"
+                            f"/{url_quote(r.entity)}"
+                            f"/{url_quote(r.project)}"
+                            f"/runs/{url_quote(r.run_id)}"
                         )
-                        print("Syncing: {} ... ".format(url), end="")  # noqa: T201
+                        print(f"Syncing: {url} ... ", end="")  # noqa: T201
                         sys.stdout.flush()
                         shown = True
             sm.finish()

wandb/util.py

@@ -36,10 +36,8 @@ from types import ModuleType
 from typing import (
     IO,
     TYPE_CHECKING,
-    Any,
     Callable,
     Dict,
-    Generator,
     Iterable,
     List,
     Mapping,
@@ -47,17 +45,12 @@ from typing import (
     Sequence,
     TextIO,
     Tuple,
-    TypeVar,
     Union,
 )
 
-if sys.version_info < (3, 10):
-    from typing_extensions import TypeGuard
-else:
-    from typing import TypeGuard
-
 import requests
 import yaml
+from typing_extensions import Any, Generator, TypeGuard, TypeVar
 
 import wandb
 import wandb.env
@@ -74,8 +67,6 @@ from wandb.sdk.lib.json_util import dump, dumps
 from wandb.sdk.lib.paths import FilePathStr, StrPath
 
 if TYPE_CHECKING:
-    import packaging.version  # type: ignore[import-not-found]
-
     import wandb.sdk.internal.settings_static
     import wandb.sdk.wandb_settings
     from wandb.sdk.artifacts.artifact import Artifact
@@ -628,9 +619,7 @@ def json_friendly(  # noqa: C901
         converted = False
     if getsizeof(obj) > VALUE_BYTES_LIMIT:
         wandb.termwarn(
-            "Serializing object of type {} that is {} bytes".format(
-                type(obj).__name__, getsizeof(obj)
-            )
+            f"Serializing object of type {type(obj).__name__} that is {getsizeof(obj)} bytes"
         )
     return obj, converted
 
@@ -1226,14 +1215,16 @@ def async_call(
         )
         thread.daemon = True
         thread.start()
+
         try:
             result = q.get(True, timeout)
-            if isinstance(result, Exception):
-                raise result.with_traceback(sys.exc_info()[2])
-            return result, thread
         except queue.Empty:
             return None, thread
 
+        if isinstance(result, Exception):
+            raise result.with_traceback(sys.exc_info()[2])
+        return result, thread
+
     return wrapper
 
 
@@ -1450,11 +1441,12 @@ def are_paths_on_same_drive(path1: str, path2: str) -> bool:
     try:
         path1_drive = pathlib.Path(path1).resolve().drive
         path2_drive = pathlib.Path(path2).resolve().drive
-        return path1_drive == path2_drive
-    # If either path is not a valid windows path, an OSError is raised.
     except OSError:
+        # If either path is not a valid Windows path, an OSError is raised.
         return False
 
+    return path1_drive == path2_drive
+
 
 # TODO(hugh): Deprecate version here and use wandb/sdk/lib/paths.py
 def to_forward_slash_path(path: str) -> str:
@@ -1548,14 +1540,19 @@ def is_unicode_safe(stream: TextIO) -> bool:
 
 
 def _has_internet() -> bool:
-    """Attempt to open a DNS connection to Googles root servers."""
+    """Returns whether we have internet access.
+
+    Checks for internet access by attempting to open a DNS connection to
+    Google's root servers.
+    """
     try:
         s = socket.create_connection(("8.8.8.8", 53), 0.5)
         s.close()
-        return True
     except OSError:
         return False
 
+    return True
+
 
 def rand_alphanumeric(
     length: int = 8, rand: Optional[Union[ModuleType, random.Random]] = None
@@ -1943,21 +1940,6 @@ def working_set() -> Iterable[InstalledDistribution]:
             pass
 
 
-def parse_version(version: str) -> "packaging.version.Version":
-    """Parse a version string into a version object.
-
-    This function is a wrapper around the `packaging.version.parse` function, which
-    is used to parse version strings into version objects. If the `packaging` library
-    is not installed, it falls back to the `pkg_resources` library.
-    """
-    try:
-        from packaging.version import parse as parse_version  # type: ignore
-    except ImportError:
-        from pkg_resources import parse_version  # type: ignore[assignment]
-
-    return parse_version(version)
-
-
 def get_core_path() -> str:
     """Returns the path to the wandb-core binary.
 

wandb/wandb_agent.py

@@ -94,14 +94,11 @@ class AgentProcess:
         if self._popen:
             # if on windows, wait() will block and we won't be able to interrupt
             if platform.system() == "Windows":
-                try:
-                    while True:
-                        p = self._popen.poll()
-                        if p is not None:
-                            return p
-                        time.sleep(1)
-                except KeyboardInterrupt:
-                    raise
+                while True:
+                    p = self._popen.poll()
+                    if p is not None:
+                        return p
+                    time.sleep(1)
             return self._popen.wait()
         return self._proc.join()
 
@@ -332,7 +329,7 @@ class Agent:
             elif command_type == "resume":
                 result = self._command_run(command)
             else:
-                raise AgentError("No such command: {}".format(command_type))
+                raise AgentError(f"No such command: {command_type}")  # noqa: TRY301
             response["result"] = result
         except Exception:
             logger.exception("Exception while processing command: %s", command)
@@ -415,7 +412,7 @@ class Agent:
                     command_list += [c]
             logger.info(
                 "About to run command: {}".format(
-                    " ".join('"{}"'.format(c) if " " in c else c for c in command_list)
+                    " ".join(f'"{c}"' if " " in c else c for c in command_list)
                 )
             )
             proc = AgentProcess(command=command_list, env=env)
@@ -466,7 +463,7 @@ class AgentApi:
 
     def command(self, command):
         command["origin"] = "local"
-        command["id"] = "local-{}".format(self._command_id)
+        command["id"] = f"local-{self._command_id}"
         self._command_id += 1
         resp_queue = self._multiproc_manager.Queue()
         command["resp_queue"] = resp_queue

{wandb-0.19.12rc1.dist-info → wandb-0.20.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: wandb
-Version: 0.19.12rc1
+Version: 0.20.1
 Summary: A CLI and library for interacting with the Weights & Biases API.
 Project-URL: Source, https://github.com/wandb/wandb
 Project-URL: Bug Reports, https://github.com/wandb/wandb/issues
@@ -48,9 +48,9 @@ Classifier: Topic :: System :: Logging
 Classifier: Topic :: System :: Monitoring
 Requires-Python: >=3.8
 Requires-Dist: click!=8.0.0,>=7.1
-Requires-Dist: docker-pycreds>=0.4.0
 Requires-Dist: eval-type-backport; python_version < '3.10'
 Requires-Dist: gitpython!=3.1.29,>=1.0.0
+Requires-Dist: packaging
 Requires-Dist: platformdirs
 Requires-Dist: protobuf!=4.21.0,!=5.28.0,<7,>=3.12.0; python_version < '3.9' and sys_platform == 'linux'
 Requires-Dist: protobuf!=4.21.0,!=5.28.0,<7,>=3.15.0; python_version == '3.9' and sys_platform == 'linux'
@@ -62,8 +62,7 @@ Requires-Dist: pyyaml
 Requires-Dist: requests<3,>=2.0.0
 Requires-Dist: sentry-sdk>=2.0.0
 Requires-Dist: setproctitle
-Requires-Dist: setuptools
-Requires-Dist: typing-extensions<5,>=4.4
+Requires-Dist: typing-extensions<5,>=4.8
 Provides-Extra: aws
 Requires-Dist: boto3; extra == 'aws'
 Requires-Dist: botocore>=1.5.76; extra == 'aws'
@@ -106,10 +105,11 @@ Requires-Dist: optuna; extra == 'launch'
 Requires-Dist: pydantic; extra == 'launch'
 Requires-Dist: pyyaml>=6.0.0; extra == 'launch'
 Requires-Dist: tomli; extra == 'launch'
+Requires-Dist: tornado>=6.5.0; (python_version >= '3.9') and extra == 'launch'
 Requires-Dist: typing-extensions; extra == 'launch'
 Provides-Extra: media
 Requires-Dist: bokeh; extra == 'media'
-Requires-Dist: imageio; extra == 'media'
+Requires-Dist: imageio>=2.28.1; extra == 'media'
 Requires-Dist: moviepy>=1.0.0; extra == 'media'
 Requires-Dist: numpy; extra == 'media'
 Requires-Dist: pillow; extra == 'media'