lightning-sdk 2025.9.2__py3-none-any.whl → 2025.9.10__py3-none-any.whl

lightning_sdk/lightning_cloud/openapi/models/v1_user_features.py

@@ -68,15 +68,18 @@ class V1UserFeatures(object):
         'f234': 'bool',
         'f236': 'bool',
         'f237': 'bool',
-        'f238': 'bool',
         'f239': 'bool',
         'f240': 'bool',
         'f241': 'bool',
-        'f242': 'bool',
         'f243': 'bool',
         'f245': 'bool',
-        'f246': 'bool',
         'f247': 'bool',
+        'f248': 'bool',
+        'f249': 'bool',
+        'f250': 'bool',
+        'f251': 'bool',
+        'f252': 'bool',
+        'f253': 'bool',
         'fair_share': 'bool',
         'featured_studios_admin': 'bool',
         'gcp_overprovisioning': 'bool',
@@ -93,7 +96,6 @@ class V1UserFeatures(object):
         'model_api_dashboard': 'bool',
         'multiple_studio_versions': 'bool',
         'nerf_fs_nonpaying': 'bool',
-        'onboarding_v2': 'bool',
         'org_level_member_permissions': 'bool',
         'org_usage_limits': 'bool',
         'persistent_disk': 'bool',
@@ -121,7 +123,6 @@ class V1UserFeatures(object):
         'studio_sharing_v2': 'bool',
         'studio_version_visibility': 'bool',
         'trainium2': 'bool',
-        'use_rclone_mounts_only': 'bool',
         'vultr': 'bool',
         'weka': 'bool',
         'writable_s3_connections': 'bool'
@@ -155,15 +156,18 @@ class V1UserFeatures(object):
         'f234': 'f234',
         'f236': 'f236',
         'f237': 'f237',
-        'f238': 'f238',
         'f239': 'f239',
         'f240': 'f240',
         'f241': 'f241',
-        'f242': 'f242',
         'f243': 'f243',
         'f245': 'f245',
-        'f246': 'f246',
         'f247': 'f247',
+        'f248': 'f248',
+        'f249': 'f249',
+        'f250': 'f250',
+        'f251': 'f251',
+        'f252': 'f252',
+        'f253': 'f253',
         'fair_share': 'fairShare',
         'featured_studios_admin': 'featuredStudiosAdmin',
         'gcp_overprovisioning': 'gcpOverprovisioning',
@@ -180,7 +184,6 @@ class V1UserFeatures(object):
         'model_api_dashboard': 'modelApiDashboard',
         'multiple_studio_versions': 'multipleStudioVersions',
         'nerf_fs_nonpaying': 'nerfFsNonpaying',
-        'onboarding_v2': 'onboardingV2',
         'org_level_member_permissions': 'orgLevelMemberPermissions',
         'org_usage_limits': 'orgUsageLimits',
         'persistent_disk': 'persistentDisk',
@@ -208,13 +211,12 @@ class V1UserFeatures(object):
         'studio_sharing_v2': 'studioSharingV2',
         'studio_version_visibility': 'studioVersionVisibility',
         'trainium2': 'trainium2',
-        'use_rclone_mounts_only': 'useRcloneMountsOnly',
         'vultr': 'vultr',
         'weka': 'weka',
         'writable_s3_connections': 'writableS3Connections'
     }
 
-    def __init__(self, affiliate_links: 'bool' =None, agents_v2: 'bool' =None, ai_hub_monetization: 'bool' =None, auto_fast_load: 'bool' =None, auto_join_orgs: 'bool' =None, b2c_experience: 'bool' =None, byo_machine_type: 'bool' =None, cap_add: 'list[str]' =None, cap_drop: 'list[str]' =None, capacity_reservation_byoc: 'bool' =None, capacity_reservation_dry_run: 'bool' =None, chat_models: 'bool' =None, cloudspace_schedules: 'bool' =None, code_tab: 'bool' =None, collab_screen_sharing: 'bool' =None, control_center_monitoring: 'bool' =None, cost_attribution_settings: 'bool' =None, custom_app_domain: 'bool' =None, datasets: 'bool' =None, default_one_cluster: 'bool' =None, deployment_persistent_disk: 'bool' =None, drive_v2: 'bool' =None, enterprise_compute_admin: 'bool' =None, f227: 'bool' =None, f234: 'bool' =None, f236: 'bool' =None, f237: 'bool' =None, f238: 'bool' =None, f239: 'bool' =None, f240: 'bool' =None, f241: 'bool' =None, f242: 'bool' =None, f243: 'bool' =None, f245: 'bool' =None, f246: 'bool' =None, f247: 'bool' =None, fair_share: 'bool' =None, featured_studios_admin: 'bool' =None, gcp_overprovisioning: 'bool' =None, gcs_connections_optimized: 'bool' =None, instant_capacity_reservation: 'bool' =None, job_artifacts_v2: 'bool' =None, kubernetes_cluster_ui: 'bool' =None, kubernetes_clusters: 'bool' =None, landing_studios: 'bool' =None, lit_logger: 'bool' =None, marketplace: 'bool' =None, mmt_fault_tolerance: 'bool' =None, mmt_strategy_selector: 'bool' =None, model_api_dashboard: 'bool' =None, multiple_studio_versions: 'bool' =None, nerf_fs_nonpaying: 'bool' =None, onboarding_v2: 'bool' =None, org_level_member_permissions: 'bool' =None, org_usage_limits: 'bool' =None, persistent_disk: 'bool' =None, plugin_distributed: 'bool' =None, plugin_inference: 'bool' =None, plugin_label_studio: 'bool' =None, plugin_langflow: 'bool' =None, plugin_python_profiler: 'bool' =None, plugin_sweeps: 'bool' =None, pricing_updates: 'bool' =None, product_generator: 'bool' =None, product_license: 'bool' =None, project_selector: 'bool' =None, publish_pipelines: 'bool' =None, reserved_machines_tab: 'bool' =None, restartable_jobs: 'bool' =None, runnable_public_studio_page: 'bool' =None, security_docs: 'bool' =None, show_dev_admin: 'bool' =None, single_wallet: 'bool' =None, slurm: 'bool' =None, specialised_studios: 'bool' =None, storage_overuse_deletion: 'bool' =None, studio_config: 'bool' =None, studio_sharing_v2: 'bool' =None, studio_version_visibility: 'bool' =None, trainium2: 'bool' =None, use_rclone_mounts_only: 'bool' =None, vultr: 'bool' =None, weka: 'bool' =None, writable_s3_connections: 'bool' =None):  # noqa: E501
+    def __init__(self, affiliate_links: 'bool' =None, agents_v2: 'bool' =None, ai_hub_monetization: 'bool' =None, auto_fast_load: 'bool' =None, auto_join_orgs: 'bool' =None, b2c_experience: 'bool' =None, byo_machine_type: 'bool' =None, cap_add: 'list[str]' =None, cap_drop: 'list[str]' =None, capacity_reservation_byoc: 'bool' =None, capacity_reservation_dry_run: 'bool' =None, chat_models: 'bool' =None, cloudspace_schedules: 'bool' =None, code_tab: 'bool' =None, collab_screen_sharing: 'bool' =None, control_center_monitoring: 'bool' =None, cost_attribution_settings: 'bool' =None, custom_app_domain: 'bool' =None, datasets: 'bool' =None, default_one_cluster: 'bool' =None, deployment_persistent_disk: 'bool' =None, drive_v2: 'bool' =None, enterprise_compute_admin: 'bool' =None, f227: 'bool' =None, f234: 'bool' =None, f236: 'bool' =None, f237: 'bool' =None, f239: 'bool' =None, f240: 'bool' =None, f241: 'bool' =None, f243: 'bool' =None, f245: 'bool' =None, f247: 'bool' =None, f248: 'bool' =None, f249: 'bool' =None, f250: 'bool' =None, f251: 'bool' =None, f252: 'bool' =None, f253: 'bool' =None, fair_share: 'bool' =None, featured_studios_admin: 'bool' =None, gcp_overprovisioning: 'bool' =None, gcs_connections_optimized: 'bool' =None, instant_capacity_reservation: 'bool' =None, job_artifacts_v2: 'bool' =None, kubernetes_cluster_ui: 'bool' =None, kubernetes_clusters: 'bool' =None, landing_studios: 'bool' =None, lit_logger: 'bool' =None, marketplace: 'bool' =None, mmt_fault_tolerance: 'bool' =None, mmt_strategy_selector: 'bool' =None, model_api_dashboard: 'bool' =None, multiple_studio_versions: 'bool' =None, nerf_fs_nonpaying: 'bool' =None, org_level_member_permissions: 'bool' =None, org_usage_limits: 'bool' =None, persistent_disk: 'bool' =None, plugin_distributed: 'bool' =None, plugin_inference: 'bool' =None, plugin_label_studio: 'bool' =None, plugin_langflow: 'bool' =None, plugin_python_profiler: 'bool' =None, plugin_sweeps: 'bool' =None, pricing_updates: 'bool' =None, product_generator: 'bool' =None, product_license: 'bool' =None, project_selector: 'bool' =None, publish_pipelines: 'bool' =None, reserved_machines_tab: 'bool' =None, restartable_jobs: 'bool' =None, runnable_public_studio_page: 'bool' =None, security_docs: 'bool' =None, show_dev_admin: 'bool' =None, single_wallet: 'bool' =None, slurm: 'bool' =None, specialised_studios: 'bool' =None, storage_overuse_deletion: 'bool' =None, studio_config: 'bool' =None, studio_sharing_v2: 'bool' =None, studio_version_visibility: 'bool' =None, trainium2: 'bool' =None, vultr: 'bool' =None, weka: 'bool' =None, writable_s3_connections: 'bool' =None):  # noqa: E501
         """V1UserFeatures - a model defined in Swagger"""  # noqa: E501
         self._affiliate_links = None
         self._agents_v2 = None
@@ -243,15 +245,18 @@ class V1UserFeatures(object):
         self._f234 = None
         self._f236 = None
         self._f237 = None
-        self._f238 = None
         self._f239 = None
         self._f240 = None
         self._f241 = None
-        self._f242 = None
         self._f243 = None
         self._f245 = None
-        self._f246 = None
         self._f247 = None
+        self._f248 = None
+        self._f249 = None
+        self._f250 = None
+        self._f251 = None
+        self._f252 = None
+        self._f253 = None
         self._fair_share = None
         self._featured_studios_admin = None
         self._gcp_overprovisioning = None
@@ -268,7 +273,6 @@ class V1UserFeatures(object):
         self._model_api_dashboard = None
         self._multiple_studio_versions = None
         self._nerf_fs_nonpaying = None
-        self._onboarding_v2 = None
         self._org_level_member_permissions = None
         self._org_usage_limits = None
         self._persistent_disk = None
@@ -296,7 +300,6 @@ class V1UserFeatures(object):
         self._studio_sharing_v2 = None
         self._studio_version_visibility = None
         self._trainium2 = None
-        self._use_rclone_mounts_only = None
         self._vultr = None
         self._weka = None
         self._writable_s3_connections = None
@@ -355,24 +358,30 @@ class V1UserFeatures(object):
             self.f236 = f236
         if f237 is not None:
             self.f237 = f237
-        if f238 is not None:
-            self.f238 = f238
         if f239 is not None:
             self.f239 = f239
         if f240 is not None:
             self.f240 = f240
         if f241 is not None:
             self.f241 = f241
-        if f242 is not None:
-            self.f242 = f242
         if f243 is not None:
             self.f243 = f243
         if f245 is not None:
             self.f245 = f245
-        if f246 is not None:
-            self.f246 = f246
         if f247 is not None:
             self.f247 = f247
+        if f248 is not None:
+            self.f248 = f248
+        if f249 is not None:
+            self.f249 = f249
+        if f250 is not None:
+            self.f250 = f250
+        if f251 is not None:
+            self.f251 = f251
+        if f252 is not None:
+            self.f252 = f252
+        if f253 is not None:
+            self.f253 = f253
         if fair_share is not None:
             self.fair_share = fair_share
         if featured_studios_admin is not None:
@@ -405,8 +414,6 @@ class V1UserFeatures(object):
             self.multiple_studio_versions = multiple_studio_versions
         if nerf_fs_nonpaying is not None:
             self.nerf_fs_nonpaying = nerf_fs_nonpaying
-        if onboarding_v2 is not None:
-            self.onboarding_v2 = onboarding_v2
         if org_level_member_permissions is not None:
             self.org_level_member_permissions = org_level_member_permissions
         if org_usage_limits is not None:
@@ -461,8 +468,6 @@ class V1UserFeatures(object):
             self.studio_version_visibility = studio_version_visibility
         if trainium2 is not None:
             self.trainium2 = trainium2
-        if use_rclone_mounts_only is not None:
-            self.use_rclone_mounts_only = use_rclone_mounts_only
         if vultr is not None:
             self.vultr = vultr
         if weka is not None:
@@ -1037,27 +1042,6 @@ class V1UserFeatures(object):
 
         self._f237 = f237
 
-    @property
-    def f238(self) -> 'bool':
-        """Gets the f238 of this V1UserFeatures.  # noqa: E501
-
-
-        :return: The f238 of this V1UserFeatures.  # noqa: E501
-        :rtype: bool
-        """
-        return self._f238
-
-    @f238.setter
-    def f238(self, f238: 'bool'):
-        """Sets the f238 of this V1UserFeatures.
-
-
-        :param f238: The f238 of this V1UserFeatures.  # noqa: E501
-        :type: bool
-        """
-
-        self._f238 = f238
-
     @property
     def f239(self) -> 'bool':
         """Gets the f239 of this V1UserFeatures.  # noqa: E501
@@ -1121,27 +1105,6 @@ class V1UserFeatures(object):
 
         self._f241 = f241
 
-    @property
-    def f242(self) -> 'bool':
-        """Gets the f242 of this V1UserFeatures.  # noqa: E501
-
-
-        :return: The f242 of this V1UserFeatures.  # noqa: E501
-        :rtype: bool
-        """
-        return self._f242
-
-    @f242.setter
-    def f242(self, f242: 'bool'):
-        """Sets the f242 of this V1UserFeatures.
-
-
-        :param f242: The f242 of this V1UserFeatures.  # noqa: E501
-        :type: bool
-        """
-
-        self._f242 = f242
-
     @property
     def f243(self) -> 'bool':
         """Gets the f243 of this V1UserFeatures.  # noqa: E501
@@ -1185,46 +1148,151 @@ class V1UserFeatures(object):
         self._f245 = f245
 
     @property
-    def f246(self) -> 'bool':
-        """Gets the f246 of this V1UserFeatures.  # noqa: E501
+    def f247(self) -> 'bool':
+        """Gets the f247 of this V1UserFeatures.  # noqa: E501
 
 
-        :return: The f246 of this V1UserFeatures.  # noqa: E501
+        :return: The f247 of this V1UserFeatures.  # noqa: E501
         :rtype: bool
         """
-        return self._f246
+        return self._f247
 
-    @f246.setter
-    def f246(self, f246: 'bool'):
-        """Sets the f246 of this V1UserFeatures.
+    @f247.setter
+    def f247(self, f247: 'bool'):
+        """Sets the f247 of this V1UserFeatures.
 
 
-        :param f246: The f246 of this V1UserFeatures.  # noqa: E501
+        :param f247: The f247 of this V1UserFeatures.  # noqa: E501
         :type: bool
         """
 
-        self._f246 = f246
+        self._f247 = f247
 
     @property
-    def f247(self) -> 'bool':
-        """Gets the f247 of this V1UserFeatures.  # noqa: E501
+    def f248(self) -> 'bool':
+        """Gets the f248 of this V1UserFeatures.  # noqa: E501
 
 
-        :return: The f247 of this V1UserFeatures.  # noqa: E501
+        :return: The f248 of this V1UserFeatures.  # noqa: E501
         :rtype: bool
         """
-        return self._f247
+        return self._f248
 
-    @f247.setter
-    def f247(self, f247: 'bool'):
-        """Sets the f247 of this V1UserFeatures.
+    @f248.setter
+    def f248(self, f248: 'bool'):
+        """Sets the f248 of this V1UserFeatures.
 
 
-        :param f247: The f247 of this V1UserFeatures.  # noqa: E501
+        :param f248: The f248 of this V1UserFeatures.  # noqa: E501
         :type: bool
         """
 
-        self._f247 = f247
+        self._f248 = f248
+
+    @property
+    def f249(self) -> 'bool':
+        """Gets the f249 of this V1UserFeatures.  # noqa: E501
+
+
+        :return: The f249 of this V1UserFeatures.  # noqa: E501
+        :rtype: bool
+        """
+        return self._f249
+
+    @f249.setter
+    def f249(self, f249: 'bool'):
+        """Sets the f249 of this V1UserFeatures.
+
+
+        :param f249: The f249 of this V1UserFeatures.  # noqa: E501
+        :type: bool
+        """
+
+        self._f249 = f249
+
+    @property
+    def f250(self) -> 'bool':
+        """Gets the f250 of this V1UserFeatures.  # noqa: E501
+
+
+        :return: The f250 of this V1UserFeatures.  # noqa: E501
+        :rtype: bool
+        """
+        return self._f250
+
+    @f250.setter
+    def f250(self, f250: 'bool'):
+        """Sets the f250 of this V1UserFeatures.
+
+
+        :param f250: The f250 of this V1UserFeatures.  # noqa: E501
+        :type: bool
+        """
+
+        self._f250 = f250
+
+    @property
+    def f251(self) -> 'bool':
+        """Gets the f251 of this V1UserFeatures.  # noqa: E501
+
+
+        :return: The f251 of this V1UserFeatures.  # noqa: E501
+        :rtype: bool
+        """
+        return self._f251
+
+    @f251.setter
+    def f251(self, f251: 'bool'):
+        """Sets the f251 of this V1UserFeatures.
+
+
+        :param f251: The f251 of this V1UserFeatures.  # noqa: E501
+        :type: bool
+        """
+
+        self._f251 = f251
+
+    @property
+    def f252(self) -> 'bool':
+        """Gets the f252 of this V1UserFeatures.  # noqa: E501
+
+
+        :return: The f252 of this V1UserFeatures.  # noqa: E501
+        :rtype: bool
+        """
+        return self._f252
+
+    @f252.setter
+    def f252(self, f252: 'bool'):
+        """Sets the f252 of this V1UserFeatures.
+
+
+        :param f252: The f252 of this V1UserFeatures.  # noqa: E501
+        :type: bool
+        """
+
+        self._f252 = f252
+
+    @property
+    def f253(self) -> 'bool':
+        """Gets the f253 of this V1UserFeatures.  # noqa: E501
+
+
+        :return: The f253 of this V1UserFeatures.  # noqa: E501
+        :rtype: bool
+        """
+        return self._f253
+
+    @f253.setter
+    def f253(self, f253: 'bool'):
+        """Sets the f253 of this V1UserFeatures.
+
+
+        :param f253: The f253 of this V1UserFeatures.  # noqa: E501
+        :type: bool
+        """
+
+        self._f253 = f253
 
     @property
     def fair_share(self) -> 'bool':
@@ -1562,27 +1630,6 @@ class V1UserFeatures(object):
 
         self._nerf_fs_nonpaying = nerf_fs_nonpaying
 
-    @property
-    def onboarding_v2(self) -> 'bool':
-        """Gets the onboarding_v2 of this V1UserFeatures.  # noqa: E501
-
-
-        :return: The onboarding_v2 of this V1UserFeatures.  # noqa: E501
-        :rtype: bool
-        """
-        return self._onboarding_v2
-
-    @onboarding_v2.setter
-    def onboarding_v2(self, onboarding_v2: 'bool'):
-        """Sets the onboarding_v2 of this V1UserFeatures.
-
-
-        :param onboarding_v2: The onboarding_v2 of this V1UserFeatures.  # noqa: E501
-        :type: bool
-        """
-
-        self._onboarding_v2 = onboarding_v2
-
     @property
     def org_level_member_permissions(self) -> 'bool':
         """Gets the org_level_member_permissions of this V1UserFeatures.  # noqa: E501
@@ -2150,27 +2197,6 @@ class V1UserFeatures(object):
 
         self._trainium2 = trainium2
 
-    @property
-    def use_rclone_mounts_only(self) -> 'bool':
-        """Gets the use_rclone_mounts_only of this V1UserFeatures.  # noqa: E501
-
-
-        :return: The use_rclone_mounts_only of this V1UserFeatures.  # noqa: E501
-        :rtype: bool
-        """
-        return self._use_rclone_mounts_only
-
-    @use_rclone_mounts_only.setter
-    def use_rclone_mounts_only(self, use_rclone_mounts_only: 'bool'):
-        """Sets the use_rclone_mounts_only of this V1UserFeatures.
-
-
-        :param use_rclone_mounts_only: The use_rclone_mounts_only of this V1UserFeatures.  # noqa: E501
-        :type: bool
-        """
-
-        self._use_rclone_mounts_only = use_rclone_mounts_only
-
     @property
     def vultr(self) -> 'bool':
         """Gets the vultr of this V1UserFeatures.  # noqa: E501

lightning_sdk/llm/llm.py

@@ -1,5 +1,6 @@
 import os
-from typing import Any, AsyncGenerator, ClassVar, Dict, Generator, List, Optional, Tuple, Union
+from dataclasses import dataclass
+from typing import Any, AsyncGenerator, ClassVar, Dict, Generator, List, Literal, Optional, Tuple, Union
 
 from lightning_sdk.api import TeamspaceApi, UserApi
 from lightning_sdk.api.llm_api import LLMApi, authenticate
@@ -15,6 +16,38 @@ PUBLIC_MODEL_PROVIDERS: Dict[str, str] = {
 }
 
 
+@dataclass
+class ModelMetadata:
+    name: str
+    provider: str
+    status: str
+    context_length: int
+    max_completion_tokens: Optional[int]
+    prompt_price: float
+    completion_price: float
+    capabilities: Dict[str, bool]
+    throughput: float
+    time_to_first_token: float
+
+    def __str__(self) -> str:
+        """Return a user-friendly string representation of the model metadata.
+
+        Returns:
+            str: A formatted multi-line string containing model information including
+                 name, provider, status, context length, pricing, performance metrics,
+                 and key capabilities (images and files support).
+        """
+        return f"""
+Model: {self.name}
+Provider: {self.provider}
+Status: {self.status}
+Context Length: {self.context_length:,} tokens
+Pricing: ${self.prompt_price:.2e}/prompt token, ${self.completion_price:.2e}/completion token
+Performance: {self.throughput:.1f} tokens/sec, {self.time_to_first_token:.1f}ms TTFT
+Capabilities: Images={self.capabilities.get('images', False)}, Files={self.capabilities.get('files', False)}
+        """.strip()
+
+
 class LLM:
     _auth_info_cached: ClassVar[bool] = False
     _cached_auth_info: ClassVar[Dict[str, Optional[str]]] = {}
@@ -70,6 +103,7 @@ class LLM:
         self._context_length = None
         self._model_id = self._get_model_id()
         self._conversations = {}
+        self._metadata = None
 
     @property
     def name(self) -> str:
@@ -79,15 +113,49 @@ class LLM:
     def provider(self) -> str:
         return self._model_provider
 
-    def context_length(self, model: Optional[str] = None) -> Optional[int]:
-        if model is None:
-            return self._context_length
+    @property
+    def metadata(self) -> ModelMetadata:
+        if self._metadata is None:
+            model = self._llm_api.get_model_metadata(self._teamspace_id, self._model_name)
+            self._metadata = ModelMetadata(
+                name=self._model_name,
+                provider=self._model_provider,
+                status=model.status,
+                context_length=int(model.context_length),
+                max_completion_tokens=int(model.max_completion_tokens) if model.max_completion_tokens != "0" else None,
+                prompt_price=model.prompt_token_price,
+                completion_price=model.completion_token_price,
+                capabilities={
+                    "images": model.abilities.can_receive_images,
+                    "files": model.abilities.can_receive_files,
+                    "hub_deployment": model.abilities.can_call_hub_deployment,
+                },
+                throughput=model.throughput,
+                time_to_first_token=model.time_to_first_token,
+            )
+        return self._metadata
 
+    @property
+    def context_length(self) -> Optional[int]:
+        """Context length for the current model."""
+        if self._context_length is None:
+            try:
+                self._context_length = self.metadata.context_length
+            except Exception as e:
+                raise ValueError(f"Cannot access context length: {e}") from e
+        return self._context_length
+
+    def get_context_length(self, model: Optional[str] = None) -> Optional[int]:
+        """Get context length for the given model."""
         context_info = self._public_assistants.get(model)
-        if context_info is None or "context_length" not in context_info:
-            raise ValueError(f"Cannot access context length of model '{model}'.")
+        if context_info and "context_length" in context_info:
+            return int(context_info["context_length"])
 
-        return int(context_info["context_length"])
+        try:
+            temp_metadata = self._llm_api.get_model_metadata(self._teamspace_id, model)
+            return int(temp_metadata.context_length)
+        except Exception as e:
+            raise ValueError(f"Cannot access context length of model '{model}': {e}") from e
 
     def _get_auth_info(self, teamspace_owner: Optional[str] = None, teamspace_name: Optional[str] = None) -> None:
         if not LLM._auth_info_cached:
@@ -291,6 +359,7 @@ class LLM:
         metadata: Optional[Dict[str, str]] = None,
         stream: bool = False,
         full_response: bool = False,
+        reasoning_effort: Optional[Literal["low", "medium", "highc"]] = None,
         **kwargs: Any,
     ) -> Union[str, AsyncGenerator[str, None]]:
         conversation_id = self._conversations.get(conversation) if conversation else None
@@ -326,10 +395,14 @@ class LLM:
         stream: bool = False,
         full_response: bool = False,
         tools: Optional[List[Dict[str, Any]]] = None,
+        reasoning_effort: Optional[Literal["low", "medium", "high"]] = None,
         **kwargs: Any,
     ) -> Union[
         V1ConversationResponseChunk, Generator[V1ConversationResponseChunk, None, None], str, Generator[str, None, None]
     ]:
+        if reasoning_effort is not None and reasoning_effort not in ["low", "medium", "high"]:
+            raise ValueError("reasoning_effort must be 'low', 'medium', 'high', or None")
+
         if conversation and conversation not in self._conversations:
             self._get_conversations()
 
@@ -352,6 +425,7 @@ class LLM:
                 metadata,
                 stream,
                 full_response,
+                reasoning_effort,
                 **kwargs,
             )
 
@@ -367,6 +441,7 @@ class LLM:
             name=conversation,
             stream=stream,
             tools=tools,
+            reasoning_effort=reasoning_effort,
             **kwargs,
         )
         if not stream:

lightning_sdk/llm/public_assistants.py

@@ -21,11 +21,11 @@ PUBLIC_MODELS = {
     },
     "google/gemini-2.5-flash": {
         "id": "ast_01jz3thxskg4fcdk4xhkjkym5a",
-        "context_length": 8000
+        "context_length": 1048576
     },
     "google/gemini-2.5-flash-lite-preview-06-17": {
         "id": "ast_01jz3thxskg4fcdk4xhkjkym5b",
-        "context_length": 8000
+        "context_length": 1048576
     },
     "lightning-ai/DeepSeek-V3.1": {
         "id": "ast_01k378z15k0msktaxjpfrfwqqh",