anyscale 0.26.22__py3-none-any.whl → 0.26.24__py3-none-any.whl

anyscale/sdk/anyscale_client/models/cluster_services_urls.py

@@ -39,6 +39,7 @@ class ClusterServicesUrls(object):
         'train_metrics_dashboard_url': 'str',
         'serve_metrics_dashboard_url': 'str',
         'serve_deployment_metrics_dashboard_url': 'str',
+        'serve_llm_metrics_dashboard_url': 'str',
         'persistent_metrics_url': 'str',
         'connect_url': 'str',
         'jupyter_notebook_url': 'str',
@@ -54,6 +55,7 @@ class ClusterServicesUrls(object):
         'train_metrics_dashboard_url': 'train_metrics_dashboard_url',
         'serve_metrics_dashboard_url': 'serve_metrics_dashboard_url',
         'serve_deployment_metrics_dashboard_url': 'serve_deployment_metrics_dashboard_url',
+        'serve_llm_metrics_dashboard_url': 'serve_llm_metrics_dashboard_url',
         'persistent_metrics_url': 'persistent_metrics_url',
         'connect_url': 'connect_url',
         'jupyter_notebook_url': 'jupyter_notebook_url',
@@ -62,7 +64,7 @@ class ClusterServicesUrls(object):
         'user_service_url': 'user_service_url'
     }
 
-    def __init__(self, webterminal_auth_url=None, metrics_dashboard_url=None, data_metrics_dashboard_url=None, train_metrics_dashboard_url=None, serve_metrics_dashboard_url=None, serve_deployment_metrics_dashboard_url=None, persistent_metrics_url=None, connect_url=None, jupyter_notebook_url=None, ray_dashboard_url=None, service_proxy_url=None, user_service_url=None, local_vars_configuration=None):  # noqa: E501
+    def __init__(self, webterminal_auth_url=None, metrics_dashboard_url=None, data_metrics_dashboard_url=None, train_metrics_dashboard_url=None, serve_metrics_dashboard_url=None, serve_deployment_metrics_dashboard_url=None, serve_llm_metrics_dashboard_url=None, persistent_metrics_url=None, connect_url=None, jupyter_notebook_url=None, ray_dashboard_url=None, service_proxy_url=None, user_service_url=None, local_vars_configuration=None):  # noqa: E501
         """ClusterServicesUrls - a model defined in OpenAPI"""  # noqa: E501
         if local_vars_configuration is None:
             local_vars_configuration = Configuration()
@@ -74,6 +76,7 @@ class ClusterServicesUrls(object):
         self._train_metrics_dashboard_url = None
         self._serve_metrics_dashboard_url = None
         self._serve_deployment_metrics_dashboard_url = None
+        self._serve_llm_metrics_dashboard_url = None
         self._persistent_metrics_url = None
         self._connect_url = None
         self._jupyter_notebook_url = None
@@ -94,6 +97,8 @@ class ClusterServicesUrls(object):
             self.serve_metrics_dashboard_url = serve_metrics_dashboard_url
         if serve_deployment_metrics_dashboard_url is not None:
             self.serve_deployment_metrics_dashboard_url = serve_deployment_metrics_dashboard_url
+        if serve_llm_metrics_dashboard_url is not None:
+            self.serve_llm_metrics_dashboard_url = serve_llm_metrics_dashboard_url
         if persistent_metrics_url is not None:
             self.persistent_metrics_url = persistent_metrics_url
         if connect_url is not None:
@@ -245,6 +250,29 @@ class ClusterServicesUrls(object):
 
         self._serve_deployment_metrics_dashboard_url = serve_deployment_metrics_dashboard_url
 
+    @property
+    def serve_llm_metrics_dashboard_url(self):
+        """Gets the serve_llm_metrics_dashboard_url of this ClusterServicesUrls.  # noqa: E501
+
+        URL for the serve LLM Grafana dashboard.  # noqa: E501
+
+        :return: The serve_llm_metrics_dashboard_url of this ClusterServicesUrls.  # noqa: E501
+        :rtype: str
+        """
+        return self._serve_llm_metrics_dashboard_url
+
+    @serve_llm_metrics_dashboard_url.setter
+    def serve_llm_metrics_dashboard_url(self, serve_llm_metrics_dashboard_url):
+        """Sets the serve_llm_metrics_dashboard_url of this ClusterServicesUrls.
+
+        URL for the serve LLM Grafana dashboard.  # noqa: E501
+
+        :param serve_llm_metrics_dashboard_url: The serve_llm_metrics_dashboard_url of this ClusterServicesUrls.  # noqa: E501
+        :type: str
+        """
+
+        self._serve_llm_metrics_dashboard_url = serve_llm_metrics_dashboard_url
+
     @property
     def persistent_metrics_url(self):
         """Gets the persistent_metrics_url of this ClusterServicesUrls.  # noqa: E501

anyscale/sdk/anyscale_client/models/compute_template_config.py

@@ -34,6 +34,8 @@ class ComputeTemplateConfig(object):
     """
     openapi_types = {
         'cloud_id': 'str',
+        'maximum_uptime_minutes': 'int',
+        'deployment_configs': 'list[CloudDeploymentComputeConfig]',
         'max_workers': 'int',
         'region': 'str',
         'allowed_azs': 'list[str]',
@@ -42,7 +44,6 @@ class ComputeTemplateConfig(object):
         'aws_advanced_configurations_json': 'object',
         'gcp_advanced_configurations_json': 'object',
         'advanced_configurations_json': 'object',
-        'maximum_uptime_minutes': 'int',
         'auto_select_worker_config': 'bool',
         'flags': 'object',
         'idle_termination_minutes': 'int'
@@ -50,6 +51,8 @@ class ComputeTemplateConfig(object):
 
     attribute_map = {
         'cloud_id': 'cloud_id',
+        'maximum_uptime_minutes': 'maximum_uptime_minutes',
+        'deployment_configs': 'deployment_configs',
         'max_workers': 'max_workers',
         'region': 'region',
         'allowed_azs': 'allowed_azs',
@@ -58,19 +61,20 @@ class ComputeTemplateConfig(object):
         'aws_advanced_configurations_json': 'aws_advanced_configurations_json',
         'gcp_advanced_configurations_json': 'gcp_advanced_configurations_json',
         'advanced_configurations_json': 'advanced_configurations_json',
-        'maximum_uptime_minutes': 'maximum_uptime_minutes',
         'auto_select_worker_config': 'auto_select_worker_config',
         'flags': 'flags',
         'idle_termination_minutes': 'idle_termination_minutes'
     }
 
-    def __init__(self, cloud_id=None, max_workers=None, region=None, allowed_azs=None, head_node_type=None, worker_node_types=None, aws_advanced_configurations_json=None, gcp_advanced_configurations_json=None, advanced_configurations_json=None, maximum_uptime_minutes=None, auto_select_worker_config=False, flags=None, idle_termination_minutes=None, local_vars_configuration=None):  # noqa: E501
+    def __init__(self, cloud_id=None, maximum_uptime_minutes=None, deployment_configs=None, max_workers=None, region=None, allowed_azs=None, head_node_type=None, worker_node_types=None, aws_advanced_configurations_json=None, gcp_advanced_configurations_json=None, advanced_configurations_json=None, auto_select_worker_config=False, flags=None, idle_termination_minutes=None, local_vars_configuration=None):  # noqa: E501
         """ComputeTemplateConfig - a model defined in OpenAPI"""  # noqa: E501
         if local_vars_configuration is None:
             local_vars_configuration = Configuration()
         self.local_vars_configuration = local_vars_configuration
 
         self._cloud_id = None
+        self._maximum_uptime_minutes = None
+        self._deployment_configs = None
         self._max_workers = None
         self._region = None
         self._allowed_azs = None
@@ -79,16 +83,20 @@ class ComputeTemplateConfig(object):
         self._aws_advanced_configurations_json = None
         self._gcp_advanced_configurations_json = None
         self._advanced_configurations_json = None
-        self._maximum_uptime_minutes = None
         self._auto_select_worker_config = None
         self._flags = None
         self._idle_termination_minutes = None
         self.discriminator = None
 
         self.cloud_id = cloud_id
+        if maximum_uptime_minutes is not None:
+            self.maximum_uptime_minutes = maximum_uptime_minutes
+        if deployment_configs is not None:
+            self.deployment_configs = deployment_configs
         if max_workers is not None:
             self.max_workers = max_workers
-        self.region = region
+        if region is not None:
+            self.region = region
         if allowed_azs is not None:
             self.allowed_azs = allowed_azs
         self.head_node_type = head_node_type
@@ -100,8 +108,6 @@ class ComputeTemplateConfig(object):
             self.gcp_advanced_configurations_json = gcp_advanced_configurations_json
         if advanced_configurations_json is not None:
             self.advanced_configurations_json = advanced_configurations_json
-        if maximum_uptime_minutes is not None:
-            self.maximum_uptime_minutes = maximum_uptime_minutes
         if auto_select_worker_config is not None:
             self.auto_select_worker_config = auto_select_worker_config
         if flags is not None:
@@ -134,11 +140,57 @@ class ComputeTemplateConfig(object):
 
         self._cloud_id = cloud_id
 
+    @property
+    def maximum_uptime_minutes(self):
+        """Gets the maximum_uptime_minutes of this ComputeTemplateConfig.  # noqa: E501
+
+        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
+
+        :return: The maximum_uptime_minutes of this ComputeTemplateConfig.  # noqa: E501
+        :rtype: int
+        """
+        return self._maximum_uptime_minutes
+
+    @maximum_uptime_minutes.setter
+    def maximum_uptime_minutes(self, maximum_uptime_minutes):
+        """Sets the maximum_uptime_minutes of this ComputeTemplateConfig.
+
+        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
+
+        :param maximum_uptime_minutes: The maximum_uptime_minutes of this ComputeTemplateConfig.  # noqa: E501
+        :type: int
+        """
+
+        self._maximum_uptime_minutes = maximum_uptime_minutes
+
+    @property
+    def deployment_configs(self):
+        """Gets the deployment_configs of this ComputeTemplateConfig.  # noqa: E501
+
+        A list of cloud deployment-specific configs to use.  # noqa: E501
+
+        :return: The deployment_configs of this ComputeTemplateConfig.  # noqa: E501
+        :rtype: list[CloudDeploymentComputeConfig]
+        """
+        return self._deployment_configs
+
+    @deployment_configs.setter
+    def deployment_configs(self, deployment_configs):
+        """Sets the deployment_configs of this ComputeTemplateConfig.
+
+        A list of cloud deployment-specific configs to use.  # noqa: E501
+
+        :param deployment_configs: The deployment_configs of this ComputeTemplateConfig.  # noqa: E501
+        :type: list[CloudDeploymentComputeConfig]
+        """
+
+        self._deployment_configs = deployment_configs
+
     @property
     def max_workers(self):
         """Gets the max_workers of this ComputeTemplateConfig.  # noqa: E501
 
-        Desired limit on total running workers for this session.  # noqa: E501
+        DEPRECATED. This attribute will be ignored - please use the \"max_resources\" flag to configure resource limits.  # noqa: E501
 
         :return: The max_workers of this ComputeTemplateConfig.  # noqa: E501
         :rtype: int
@@ -149,7 +201,7 @@ class ComputeTemplateConfig(object):
     def max_workers(self, max_workers):
         """Sets the max_workers of this ComputeTemplateConfig.
 
-        Desired limit on total running workers for this session.  # noqa: E501
+        DEPRECATED. This attribute will be ignored - please use the \"max_resources\" flag to configure resource limits.  # noqa: E501
 
         :param max_workers: The max_workers of this ComputeTemplateConfig.  # noqa: E501
         :type: int
@@ -177,8 +229,6 @@ class ComputeTemplateConfig(object):
         :param region: The region of this ComputeTemplateConfig.  # noqa: E501
         :type: str
         """
-        if self.local_vars_configuration.client_side_validation and region is None:  # noqa: E501
-            raise ValueError("Invalid value for `region`, must not be `None`")  # noqa: E501
 
         self._region = region
 
@@ -322,29 +372,6 @@ class ComputeTemplateConfig(object):
 
         self._advanced_configurations_json = advanced_configurations_json
 
-    @property
-    def maximum_uptime_minutes(self):
-        """Gets the maximum_uptime_minutes of this ComputeTemplateConfig.  # noqa: E501
-
-        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
-
-        :return: The maximum_uptime_minutes of this ComputeTemplateConfig.  # noqa: E501
-        :rtype: int
-        """
-        return self._maximum_uptime_minutes
-
-    @maximum_uptime_minutes.setter
-    def maximum_uptime_minutes(self, maximum_uptime_minutes):
-        """Sets the maximum_uptime_minutes of this ComputeTemplateConfig.
-
-        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
-
-        :param maximum_uptime_minutes: The maximum_uptime_minutes of this ComputeTemplateConfig.  # noqa: E501
-        :type: int
-        """
-
-        self._maximum_uptime_minutes = maximum_uptime_minutes
-
     @property
     def auto_select_worker_config(self):
         """Gets the auto_select_worker_config of this ComputeTemplateConfig.  # noqa: E501

anyscale/sdk/anyscale_client/models/create_cluster_compute_config.py

@@ -34,6 +34,8 @@ class CreateClusterComputeConfig(object):
     """
     openapi_types = {
         'cloud_id': 'str',
+        'maximum_uptime_minutes': 'int',
+        'deployment_configs': 'list[CloudDeploymentComputeConfig]',
         'max_workers': 'int',
         'region': 'str',
         'allowed_azs': 'list[str]',
@@ -42,7 +44,6 @@ class CreateClusterComputeConfig(object):
         'aws_advanced_configurations_json': 'object',
         'gcp_advanced_configurations_json': 'object',
         'advanced_configurations_json': 'object',
-        'maximum_uptime_minutes': 'int',
         'auto_select_worker_config': 'bool',
         'flags': 'object',
         'idle_termination_minutes': 'int'
@@ -50,6 +51,8 @@ class CreateClusterComputeConfig(object):
 
     attribute_map = {
         'cloud_id': 'cloud_id',
+        'maximum_uptime_minutes': 'maximum_uptime_minutes',
+        'deployment_configs': 'deployment_configs',
         'max_workers': 'max_workers',
         'region': 'region',
         'allowed_azs': 'allowed_azs',
@@ -58,19 +61,20 @@ class CreateClusterComputeConfig(object):
         'aws_advanced_configurations_json': 'aws_advanced_configurations_json',
         'gcp_advanced_configurations_json': 'gcp_advanced_configurations_json',
         'advanced_configurations_json': 'advanced_configurations_json',
-        'maximum_uptime_minutes': 'maximum_uptime_minutes',
         'auto_select_worker_config': 'auto_select_worker_config',
         'flags': 'flags',
         'idle_termination_minutes': 'idle_termination_minutes'
     }
 
-    def __init__(self, cloud_id=None, max_workers=None, region='USE_CLOUD', allowed_azs=None, head_node_type=None, worker_node_types=None, aws_advanced_configurations_json=None, gcp_advanced_configurations_json=None, advanced_configurations_json=None, maximum_uptime_minutes=None, auto_select_worker_config=False, flags=None, idle_termination_minutes=120, local_vars_configuration=None):  # noqa: E501
+    def __init__(self, cloud_id=None, maximum_uptime_minutes=None, deployment_configs=None, max_workers=None, region='USE_CLOUD', allowed_azs=None, head_node_type=None, worker_node_types=None, aws_advanced_configurations_json=None, gcp_advanced_configurations_json=None, advanced_configurations_json=None, auto_select_worker_config=False, flags=None, idle_termination_minutes=120, local_vars_configuration=None):  # noqa: E501
         """CreateClusterComputeConfig - a model defined in OpenAPI"""  # noqa: E501
         if local_vars_configuration is None:
             local_vars_configuration = Configuration()
         self.local_vars_configuration = local_vars_configuration
 
         self._cloud_id = None
+        self._maximum_uptime_minutes = None
+        self._deployment_configs = None
         self._max_workers = None
         self._region = None
         self._allowed_azs = None
@@ -79,13 +83,16 @@ class CreateClusterComputeConfig(object):
         self._aws_advanced_configurations_json = None
         self._gcp_advanced_configurations_json = None
         self._advanced_configurations_json = None
-        self._maximum_uptime_minutes = None
         self._auto_select_worker_config = None
         self._flags = None
         self._idle_termination_minutes = None
         self.discriminator = None
 
         self.cloud_id = cloud_id
+        if maximum_uptime_minutes is not None:
+            self.maximum_uptime_minutes = maximum_uptime_minutes
+        if deployment_configs is not None:
+            self.deployment_configs = deployment_configs
         if max_workers is not None:
             self.max_workers = max_workers
         if region is not None:
@@ -101,8 +108,6 @@ class CreateClusterComputeConfig(object):
             self.gcp_advanced_configurations_json = gcp_advanced_configurations_json
         if advanced_configurations_json is not None:
             self.advanced_configurations_json = advanced_configurations_json
-        if maximum_uptime_minutes is not None:
-            self.maximum_uptime_minutes = maximum_uptime_minutes
         if auto_select_worker_config is not None:
             self.auto_select_worker_config = auto_select_worker_config
         if flags is not None:
@@ -135,11 +140,57 @@ class CreateClusterComputeConfig(object):
 
         self._cloud_id = cloud_id
 
+    @property
+    def maximum_uptime_minutes(self):
+        """Gets the maximum_uptime_minutes of this CreateClusterComputeConfig.  # noqa: E501
+
+        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
+
+        :return: The maximum_uptime_minutes of this CreateClusterComputeConfig.  # noqa: E501
+        :rtype: int
+        """
+        return self._maximum_uptime_minutes
+
+    @maximum_uptime_minutes.setter
+    def maximum_uptime_minutes(self, maximum_uptime_minutes):
+        """Sets the maximum_uptime_minutes of this CreateClusterComputeConfig.
+
+        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
+
+        :param maximum_uptime_minutes: The maximum_uptime_minutes of this CreateClusterComputeConfig.  # noqa: E501
+        :type: int
+        """
+
+        self._maximum_uptime_minutes = maximum_uptime_minutes
+
+    @property
+    def deployment_configs(self):
+        """Gets the deployment_configs of this CreateClusterComputeConfig.  # noqa: E501
+
+        A list of cloud deployment-specific configs to use.  # noqa: E501
+
+        :return: The deployment_configs of this CreateClusterComputeConfig.  # noqa: E501
+        :rtype: list[CloudDeploymentComputeConfig]
+        """
+        return self._deployment_configs
+
+    @deployment_configs.setter
+    def deployment_configs(self, deployment_configs):
+        """Sets the deployment_configs of this CreateClusterComputeConfig.
+
+        A list of cloud deployment-specific configs to use.  # noqa: E501
+
+        :param deployment_configs: The deployment_configs of this CreateClusterComputeConfig.  # noqa: E501
+        :type: list[CloudDeploymentComputeConfig]
+        """
+
+        self._deployment_configs = deployment_configs
+
     @property
     def max_workers(self):
         """Gets the max_workers of this CreateClusterComputeConfig.  # noqa: E501
 
-        Desired limit on total running workers for this Cluster.  # noqa: E501
+        DEPRECATED. This attribute will be ignored - please use the \"max_resources\" flag to configure resource limits.  # noqa: E501
 
         :return: The max_workers of this CreateClusterComputeConfig.  # noqa: E501
         :rtype: int
@@ -150,7 +201,7 @@ class CreateClusterComputeConfig(object):
     def max_workers(self, max_workers):
         """Sets the max_workers of this CreateClusterComputeConfig.
 
-        Desired limit on total running workers for this Cluster.  # noqa: E501
+        DEPRECATED. This attribute will be ignored - please use the \"max_resources\" flag to configure resource limits.  # noqa: E501
 
         :param max_workers: The max_workers of this CreateClusterComputeConfig.  # noqa: E501
         :type: int
@@ -321,29 +372,6 @@ class CreateClusterComputeConfig(object):
 
         self._advanced_configurations_json = advanced_configurations_json
 
-    @property
-    def maximum_uptime_minutes(self):
-        """Gets the maximum_uptime_minutes of this CreateClusterComputeConfig.  # noqa: E501
-
-        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
-
-        :return: The maximum_uptime_minutes of this CreateClusterComputeConfig.  # noqa: E501
-        :rtype: int
-        """
-        return self._maximum_uptime_minutes
-
-    @maximum_uptime_minutes.setter
-    def maximum_uptime_minutes(self, maximum_uptime_minutes):
-        """Sets the maximum_uptime_minutes of this CreateClusterComputeConfig.
-
-        If set to a positive number, Anyscale will terminate the cluster this many minutes after cluster start.  # noqa: E501
-
-        :param maximum_uptime_minutes: The maximum_uptime_minutes of this CreateClusterComputeConfig.  # noqa: E501
-        :type: int
-        """
-
-        self._maximum_uptime_minutes = maximum_uptime_minutes
-
     @property
     def auto_select_worker_config(self):
         """Gets the auto_select_worker_config of this CreateClusterComputeConfig.  # noqa: E501

anyscale/sdk/anyscale_client/models/session.py

@@ -60,6 +60,7 @@ class Session(object):
         'train_metrics_dashboard_url': 'str',
         'serve_metrics_dashboard_url': 'str',
         'serve_deployment_metrics_dashboard_url': 'str',
+        'serve_llm_metrics_dashboard_url': 'str',
         'persistent_metrics_url': 'str',
         'connect_url': 'str',
         'jupyter_notebook_url': 'str',
@@ -114,6 +115,7 @@ class Session(object):
         'train_metrics_dashboard_url': 'train_metrics_dashboard_url',
         'serve_metrics_dashboard_url': 'serve_metrics_dashboard_url',
         'serve_deployment_metrics_dashboard_url': 'serve_deployment_metrics_dashboard_url',
+        'serve_llm_metrics_dashboard_url': 'serve_llm_metrics_dashboard_url',
         'persistent_metrics_url': 'persistent_metrics_url',
         'connect_url': 'connect_url',
         'jupyter_notebook_url': 'jupyter_notebook_url',
@@ -140,7 +142,7 @@ class Session(object):
         'ray_dashboard_snapshot_last_reported_at': 'ray_dashboard_snapshot_last_reported_at'
     }
 
-    def __init__(self, name=None, project_id=None, cloud_id=None, cluster_config=None, build_id=None, compute_template_id=None, idle_timeout=120, uses_app_config=False, allow_public_internet_traffic=False, user_service_access=None, user_service_token=None, ha_job_id=None, id=None, state=None, pending_state=None, state_data=None, status=None, status_details=None, creator_id=None, created_at=None, archived_at=None, webterminal_auth_url=None, metrics_dashboard_url=None, data_metrics_dashboard_url=None, train_metrics_dashboard_url=None, serve_metrics_dashboard_url=None, serve_deployment_metrics_dashboard_url=None, persistent_metrics_url=None, connect_url=None, jupyter_notebook_url=None, ray_dashboard_url=None, access_token=None, service_proxy_url=None, tensorboard_available=None, cluster_config_last_modified_at=None, host_name=None, head_node_ip=None, ssh_authorized_keys=None, ssh_private_key=None, anyscaled_config=None, anyscaled_config_generated_at=None, default_build_id=None, idle_timeout_last_activity_at=None, ray_version=None, ray_version_last_updated_at=None, user_service_url=None, ray_component_activities_last_reported_at=None, activity_details=None, maximum_uptime_will_terminate_cluster_at=None, idle_termination_status=None, ray_dashboard_snapshot_last_reported_at=None, local_vars_configuration=None):  # noqa: E501
+    def __init__(self, name=None, project_id=None, cloud_id=None, cluster_config=None, build_id=None, compute_template_id=None, idle_timeout=120, uses_app_config=False, allow_public_internet_traffic=False, user_service_access=None, user_service_token=None, ha_job_id=None, id=None, state=None, pending_state=None, state_data=None, status=None, status_details=None, creator_id=None, created_at=None, archived_at=None, webterminal_auth_url=None, metrics_dashboard_url=None, data_metrics_dashboard_url=None, train_metrics_dashboard_url=None, serve_metrics_dashboard_url=None, serve_deployment_metrics_dashboard_url=None, serve_llm_metrics_dashboard_url=None, persistent_metrics_url=None, connect_url=None, jupyter_notebook_url=None, ray_dashboard_url=None, access_token=None, service_proxy_url=None, tensorboard_available=None, cluster_config_last_modified_at=None, host_name=None, head_node_ip=None, ssh_authorized_keys=None, ssh_private_key=None, anyscaled_config=None, anyscaled_config_generated_at=None, default_build_id=None, idle_timeout_last_activity_at=None, ray_version=None, ray_version_last_updated_at=None, user_service_url=None, ray_component_activities_last_reported_at=None, activity_details=None, maximum_uptime_will_terminate_cluster_at=None, idle_termination_status=None, ray_dashboard_snapshot_last_reported_at=None, local_vars_configuration=None):  # noqa: E501
         """Session - a model defined in OpenAPI"""  # noqa: E501
         if local_vars_configuration is None:
             local_vars_configuration = Configuration()
@@ -173,6 +175,7 @@ class Session(object):
         self._train_metrics_dashboard_url = None
         self._serve_metrics_dashboard_url = None
         self._serve_deployment_metrics_dashboard_url = None
+        self._serve_llm_metrics_dashboard_url = None
         self._persistent_metrics_url = None
         self._connect_url = None
         self._jupyter_notebook_url = None
@@ -245,6 +248,8 @@ class Session(object):
             self.serve_metrics_dashboard_url = serve_metrics_dashboard_url
         if serve_deployment_metrics_dashboard_url is not None:
             self.serve_deployment_metrics_dashboard_url = serve_deployment_metrics_dashboard_url
+        if serve_llm_metrics_dashboard_url is not None:
+            self.serve_llm_metrics_dashboard_url = serve_llm_metrics_dashboard_url
         if persistent_metrics_url is not None:
             self.persistent_metrics_url = persistent_metrics_url
         if connect_url is not None:
@@ -926,6 +931,29 @@ class Session(object):
 
         self._serve_deployment_metrics_dashboard_url = serve_deployment_metrics_dashboard_url
 
+    @property
+    def serve_llm_metrics_dashboard_url(self):
+        """Gets the serve_llm_metrics_dashboard_url of this Session.  # noqa: E501
+
+        URL for Serve LLM Grafana dashboard for this Session. This field will only be populated after the Session finishes starting.  # noqa: E501
+
+        :return: The serve_llm_metrics_dashboard_url of this Session.  # noqa: E501
+        :rtype: str
+        """
+        return self._serve_llm_metrics_dashboard_url
+
+    @serve_llm_metrics_dashboard_url.setter
+    def serve_llm_metrics_dashboard_url(self, serve_llm_metrics_dashboard_url):
+        """Sets the serve_llm_metrics_dashboard_url of this Session.
+
+        URL for Serve LLM Grafana dashboard for this Session. This field will only be populated after the Session finishes starting.  # noqa: E501
+
+        :param serve_llm_metrics_dashboard_url: The serve_llm_metrics_dashboard_url of this Session.  # noqa: E501
+        :type: str
+        """
+
+        self._serve_llm_metrics_dashboard_url = serve_llm_metrics_dashboard_url
+
     @property
     def persistent_metrics_url(self):
         """Gets the persistent_metrics_url of this Session.  # noqa: E501
@@ -1022,7 +1050,7 @@ class Session(object):
     def access_token(self):
         """Gets the access_token of this Session.  # noqa: E501
 
-        Access token for web based services (e.g. jupyter, tensorboard, etc). This field will be populated when the web based services are available after the Session finishes starting.  # noqa: E501
+        [DEPRECATED] Call GET /api/v2/authentication/{cluster_id}/cluster_access_token to get this.  # noqa: E501
 
         :return: The access_token of this Session.  # noqa: E501
         :rtype: str
@@ -1033,7 +1061,7 @@ class Session(object):
     def access_token(self, access_token):
         """Sets the access_token of this Session.
 
-        Access token for web based services (e.g. jupyter, tensorboard, etc). This field will be populated when the web based services are available after the Session finishes starting.  # noqa: E501
+        [DEPRECATED] Call GET /api/v2/authentication/{cluster_id}/cluster_access_token to get this.  # noqa: E501
 
         :param access_token: The access_token of this Session.  # noqa: E501
         :type: str