libinephany 1.1.1__tar.gz → 1.1.3__tar.gz

libinephany-1.1.3/CODE_VERSION.cfg

@@ -0,0 +1 @@
+1.1.3

{libinephany-1.1.1/libinephany.egg-info → libinephany-1.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: libinephany
-Version: 1.1.1
+Version: 1.1.3
 Summary: Inephany library containing code commonly used by multiple subpackages.
 Author-email: Inephany <info@inephany.com>
 License: Apache 2.0

{libinephany-1.1.1 → libinephany-1.1.3}/libinephany/pydantic_models/schemas/inner_task_profile.py

@@ -7,8 +7,15 @@
 import math
 from typing import Any, Callable
 
-from loguru import logger
-from pydantic import BaseModel, field_validator
+from pydantic import BaseModel
+
+# ======================================================================================================================
+#
+# CONSTANTS
+#
+# ======================================================================================================================
+
+VRAM_USAGES_KEY = "vram_usages"
 
 # ======================================================================================================================
 #
@@ -20,26 +27,40 @@ from pydantic import BaseModel, field_validator
 class InnerTaskProfile(BaseModel):
 
     inner_task_name: str
+
     number_of_agents: int
     number_of_layers: int
+    number_of_parameters: int
+
     observation_space_sizes: dict[str, int]
     action_space_sizes: dict[str, int]
-    number_of_parameters: int
-    vram_usage: float
-    idle_vram_usage: float
-    hparam_overrides: dict[str, dict[str, Any]] | None = None
 
-    @field_validator("vram_usage", "idle_vram_usage", mode="before")
-    def replace_non_with_nan(cls, value: None | float) -> float:
+    vram_usages: dict[int, tuple[float, float]] | None = None
+    expected_vram_usage: float | None = None
+    expected_idle_vram_usage: float | None = None
+    max_batch_size_override: int | None = None
+
+    @property
+    def vram_usage(self) -> float:
+        """
+        :return: VRAM usage at the max batch size.
+        """
+
+        if self.expected_vram_usage is None:
+            return float("nan")
+
+        return self.expected_vram_usage
+
+    @property
+    def idle_vram_usage(self) -> float:
         """
-        :param value: Value to replace with NaN if it is None.
-        :return: Either the given float value or NaN.
+        :return: Idle VRAM usage at the max batch size.
         """
 
-        if value is None:
+        if self.expected_idle_vram_usage is None:
             return float("nan")
 
-        return value
+        return self.expected_idle_vram_usage
 
     @property
     def failed_to_profile(self) -> bool:
@@ -49,19 +70,18 @@ class InnerTaskProfile(BaseModel):
 
         return math.isnan(self.vram_usage)
 
-    def model_dump_json(self, **kwargs) -> str:
+    def model_dump(self, **kwargs) -> dict[str, Any]:
         """
         :param kwargs: Standard Pydantic model dump kwargs.
         :return: Dump result of the superclass' method.
         """
 
-        logger.debug(
-            f"Inner task {self.inner_task_name} consumed {self.vram_usage:.3f} MB of VRAM while training and "
-            f"{self.idle_vram_usage:.3f} MB of VRAM while idle. It has {self.number_of_agents} agents across "
-            f"{self.number_of_layers} inner model layers."
-        )
+        super_dump = super().model_dump(**kwargs)
 
-        return super().model_dump_json(**kwargs)
+        if self.vram_usages is not None:
+            super_dump[VRAM_USAGES_KEY] = {k: list(v) for k, v in self.vram_usages.items()}
+
+        return super_dump
 
 
 class InnerTaskProfiles(BaseModel):
@@ -235,42 +255,6 @@ class InnerTaskProfiles(BaseModel):
 
         return inner_task_name in self.profiles
 
-    def add_profile(
-        self,
-        inner_task_name: str,
-        number_of_agents: int,
-        number_of_layers: int,
-        observation_space_sizes: dict[str, int],
-        action_space_sizes: dict[str, int],
-        number_of_parameters: int,
-        vram_usage: float,
-        idle_vram_usage: float,
-        hparam_overrides: dict[str, dict[str, Any]] | None = None,
-    ) -> None:
-        """
-        :param inner_task_name: Name of the inner task to add a profile for.
-        :param number_of_agents: Number of agents active in the inner task's environment.
-        :param number_of_layers: Number of layers in the inner model.
-        :param observation_space_sizes: Dictionary mapping agent IDs to their observation space sizes.
-        :param action_space_sizes: Dictionary mapping agent IDs to their action space sizes.
-        :param vram_usage: VRAM required to perform the inner task. Can be NaN if an OOM was encountered.
-        :param idle_vram_usage: VRAM required for the inner task to sit loaded but not actively being trained. Can be
-        NaN if an OOM was encountered.
-        :param hparam_overrides: Hyperparameter overrides for the inner task.
-        """
-
-        self.profiles[inner_task_name] = InnerTaskProfile(
-            inner_task_name=inner_task_name,
-            number_of_agents=number_of_agents,
-            number_of_layers=number_of_layers,
-            observation_space_sizes=observation_space_sizes,
-            action_space_sizes=action_space_sizes,
-            number_of_parameters=number_of_parameters,
-            vram_usage=vram_usage,
-            idle_vram_usage=idle_vram_usage,
-            hparam_overrides=hparam_overrides,
-        )
-
     def validate_task_profiles(self, policy_mapping_function: Callable[[str, Any, Any], str]) -> None:
         """
         :param policy_mapping_function: Function which maps agent IDs to policy IDs.

{libinephany-1.1.1 → libinephany-1.1.3}/libinephany/web_apps/error_logger.py

@@ -58,6 +58,8 @@ class ErrorLogger(AsyncioWorker):
     def __init__(
         self,
         service_name: str,
+        member_ids_to_mention: list[str] | None = None,
+        custom_error_message: str | None = None,
         minimum_severity: ErrorSeverities = ErrorSeverities.WARNING,
         error_classifier: ExceptionSeverityClassifier | None = None,
         max_time_accumulating_errors: float = 10.0,
@@ -66,6 +68,8 @@ class ErrorLogger(AsyncioWorker):
         """
         :param service_name: Name of the service deployed on the cloud errors sent to this error logger should be sent
         to.
+        :param member_ids_to_mention: List of member IDs to mention in the Slack message. If None, no members will be mentioned.
+        :param custom_error_message: Custom error message to include in the Slack message. If None, no custom message will be included.
         :param minimum_severity: Minimum error severity that can be sent to Slack.
         :param error_classifier: Optional callable used to alter how errors severities are classified. If this is None
         all errors are given the 'WARNING' severity.
@@ -77,6 +81,8 @@ class ErrorLogger(AsyncioWorker):
 
         self._slack_client, self._channel_id = self._get_slack_client()
 
+        self.member_mentions = self._form_member_mentions(member_ids_to_mention=member_ids_to_mention)
+        self.custom_error_message = custom_error_message
         self.service_name = service_name
         self.error_classifier = error_classifier if error_classifier is not None else default_error_classifier
         self.max_time_accumulating_errors = max_time_accumulating_errors
@@ -144,6 +150,7 @@ class ErrorLogger(AsyncioWorker):
         exception: Exception,
         frequency: int,
         frequency_time_window: float,
+        custom_error_message: str | None,
     ) -> str:
         """
         :param mention: Mention tag of the entire channel or a particular user ID.
@@ -152,19 +159,40 @@ class ErrorLogger(AsyncioWorker):
         :param exception: Exception that occurred.
         :param frequency: How frequently the exception occurred in the given time window.
         :param frequency_time_window: Time between the first and last instance of the exception.
+        :param custom_error_message: Custom error message to include in the Slack message. If None, no custom message will be included.
         :return: Formatted Slack message string.
         """
 
         frequency_line = f"{frequency} in {frequency_time_window}s" if frequency > 1 else f"{frequency}"
 
+        header = f"{notification_emoji} *{service_name} Error* {notification_emoji}\n"
+
+        if mention:
+            header += f"• *Alerting*: {mention}\n"
+
+        if custom_error_message is not None:
+            header += f"• *Custom Message*: {custom_error_message}\n"
+
         return (
-            f"{mention}{notification_emoji} *{service_name} Error* {notification_emoji}\n"
+            f"{header}"
             f"• *Service*: {service_name}\n"
             f"• *Error Type*: {exception.__class__.__name__}\n"
             f"• *Frequency*: {frequency_line}\n"
             f"• *Traceback*: "
         )
 
+    @staticmethod
+    def _form_member_mentions(member_ids_to_mention: list[str] | None) -> str:
+        """
+        :param member_ids_to_mention: List of member IDs to mention in the Slack message.
+        :return: String of member mentions.
+        """
+
+        if not member_ids_to_mention:
+            return ""
+
+        return " ".join([f"<@{member_id}>" for member_id in member_ids_to_mention])
+
     def _get_traceback_file_name(self, exception: Exception, exception_timestamp: str) -> str:
         """
         :param exception: Exception being sent to Slack.
@@ -275,7 +303,10 @@ class ErrorLogger(AsyncioWorker):
         traceback_filename = self._get_traceback_file_name(exception=exception, exception_timestamp=formatted_timestamp)
 
         notification_emoji = SEVERITY_EMOJIS[severity]
-        mention = "" if not ErrorSeverities.should_mention(severity=severity) else self.CHANNEL_MENTION
+        # Temporary since the bot is used in training and the API.
+        mention = (
+            self.member_mentions if not ErrorSeverities.should_mention(severity=severity) else self.CHANNEL_MENTION
+        )
 
         self._log_error_to_terminal(exception=exception)
 
@@ -291,6 +322,7 @@ class ErrorLogger(AsyncioWorker):
                     exception=exception,
                     frequency=frequency,
                     frequency_time_window=frequency_time_window,
+                    custom_error_message=self.custom_error_message,
                 ),
             )
         except SlackApiError as e:

{libinephany-1.1.1 → libinephany-1.1.3/libinephany.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: libinephany
-Version: 1.1.1
+Version: 1.1.3
 Summary: Inephany library containing code commonly used by multiple subpackages.
 Author-email: Inephany <info@inephany.com>
 License: Apache 2.0

libinephany-1.1.1/CODE_VERSION.cfg

@@ -1 +0,0 @@
-1.1.1