supervisely 6.73.406__py3-none-any.whl → 6.73.408__py3-none-any.whl

supervisely/api/app_api.py

@@ -11,7 +11,7 @@ from typing_extensions import Literal
 
 from supervisely._utils import is_community, is_development, take_with_default
 from supervisely.api.module_api import ApiField
-from supervisely.api.task_api import TaskApi
+from supervisely.api.task_api import KubernetesSettings, TaskApi
 
 # from supervisely.app.constants import DATA, STATE, CONTEXT, TEMPLATE
 STATE = "state"
@@ -1682,6 +1682,7 @@ class AppApi(TaskApi):
         proxy_keep_url: bool = False,
         module_id: Optional[int] = None,
         redirect_requests: Dict[str, int] = {},
+        kubernetes_settings: Optional[Union[KubernetesSettings, Dict[str, Any]]] = None,
     ) -> SessionInfo:
         """Start a new application session (task).
 
@@ -1713,6 +1714,8 @@ class AppApi(TaskApi):
         :type module_id: Optional[int]
         :param redirect_requests: For internal usage only in Develop and Debug mode.
         :type redirect_requests: dict
+        :param kubernetes_settings: Kubernetes settings for the task. If not specified, default settings will be used.
+        :type kubernetes_settings: Optional[Union[KubernetesSettings, Dict[str, Any]]]
         :return: SessionInfo object with information about the started task.
         :rtype: SessionInfo
         :raises ValueError: If both app_id and module_id are not provided.
@@ -1747,6 +1750,7 @@ class AppApi(TaskApi):
             proxy_keep_url=proxy_keep_url,
             module_id=module_id,
             redirect_requests=redirect_requests,
+            kubernetes_settings=kubernetes_settings,
         )
         if type(result) is not list:
             result = [result]

supervisely/api/task_api.py

@@ -11,6 +11,7 @@ from pathlib import Path
 from typing import Any, Callable, Dict, List, Literal, NamedTuple, Optional, Union
 
 import requests
+from pydantic import BaseModel, Field
 from requests_toolbelt import MultipartEncoder, MultipartEncoderMonitor
 from tqdm import tqdm
 
@@ -31,6 +32,28 @@ from supervisely.io.fs import (
 )
 
 
+class KubernetesSettings(BaseModel):
+    """
+    KubernetesSettings for application resource limits and requests.
+    """
+
+    use_health_check: Optional[bool] = Field(None, alias="useHealthCheck")
+    request_cpus: Optional[int] = Field(None, alias="requestCpus")
+    limit_cpus: Optional[int] = Field(None, alias="limitCpus")
+    limit_memory_gb: Optional[int] = Field(None, alias="limitMemoryGb")
+    limit_shm_gb: Optional[int] = Field(None, alias="limitShmGb")
+    limit_storage_gb: Optional[int] = Field(None, alias="limitStorageGb")
+    limit_gpus: Optional[int] = Field(None, alias="limitGpus")
+    limit_gpu_memory_mb: Optional[int] = Field(None, alias="limitGpuMemoryMb")
+    limit_gpu_cores_perc: Optional[int] = Field(None, alias="limitGpuCoresPerc")
+
+    model_config = {"populate_by_name": True}
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dict with only non-None values using aliases."""
+        return self.model_dump(exclude_none=True, by_alias=True)
+
+
 class TaskFinishedWithError(Exception):
     """TaskFinishedWithError"""
 
@@ -365,6 +388,7 @@ class TaskApi(ModuleApiBase, ModuleWithStatus):
         module_id: Optional[int] = None,
         redirect_requests: Optional[Dict[str, int]] = {},
         limit_by_workspace: bool = False,
+        kubernetes_settings: Optional[Union[KubernetesSettings, Dict[str, Any]]] = None,
     ) -> Dict[str, Any]:
         """Starts the application task on the agent.
 
@@ -401,6 +425,8 @@ class TaskApi(ModuleApiBase, ModuleWithStatus):
         :param limit_by_workspace: If set to True tasks will be only visible inside of the workspace
             with specified workspace_id.
         :type limit_by_workspace: bool, optional
+        :param kubernetes_settings: Kubernetes settings for the application.
+        :type kubernetes_settings: Union[KubernetesSettings, Dict[str, Any]], optional
         :return: Task information in JSON format.
         :rtype: Dict[str, Any]
 
@@ -439,6 +465,15 @@ class TaskApi(ModuleApiBase, ModuleWithStatus):
             ApiField.LIMIT_BY_WORKSPACE: limit_by_workspace,
         }
 
+        if kubernetes_settings is not None:
+            if isinstance(kubernetes_settings, KubernetesSettings):
+                kubernetes_settings = kubernetes_settings.to_dict()
+            if not isinstance(kubernetes_settings, dict):
+                raise TypeError(
+                    f"kubernetes_settings must be a dict or an instance of KubernetesSettings, got {type(kubernetes_settings)}"
+                )
+            advanced_settings.update(kubernetes_settings)
+
         data = {
             ApiField.AGENT_ID: agent_id,
             # "nodeId": agent_id,

supervisely/nn/inference/inference.py

@@ -182,6 +182,9 @@ class Inference:
         self.classes: List[str] = None
         self._model_dir = model_dir
         self._model_served = False
+        self._freeze_timer = None
+        self._model_frozen = False
+        self._inactivity_timeout = 3600  # 1 hour
         self._deploy_params: dict = None
         self._model_meta = None
         self._confidence = "confidence"
@@ -225,6 +228,7 @@ class Inference:
                 deploy_params["model_files"] = local_model_files
             logger.debug("Loading model...")
             self._load_model_headless(**deploy_params)
+            self._schedule_freeze_on_inactivity()
 
         if self._use_gui:
             initialize_custom_gui_method = getattr(self, "initialize_custom_gui", None)
@@ -273,6 +277,7 @@ class Inference:
                     self.device = device
                     self.load_on_device(self._model_dir, device)
                     gui.show_deployed_model_info(self)
+                self._schedule_freeze_on_inactivity()
 
             def on_change_model_callback(
                 gui: Union[GUI.InferenceGUI, GUI.ServingGUI, GUI.ServingGUITemplate],
@@ -407,6 +412,7 @@ class Inference:
         deploy_params = self._get_deploy_parameters_from_custom_checkpoint(checkpoint_path, device, runtime)
         logger.debug(f"Deploying custom model '{checkpoint_name}'...")
         self._load_model_headless(**deploy_params)
+        self._schedule_freeze_on_inactivity()
         return self
 
     def get_batch_size(self):
@@ -1228,6 +1234,10 @@ class Inference:
 
     def shutdown_model(self):
         self._model_served = False
+        self._model_frozen = False
+        if self._freeze_timer is not None:
+            self._freeze_timer.cancel()
+            self._freeze_timer = None
         self.device = None
         self.runtime = None
         self.model_precision = None
@@ -1437,12 +1447,13 @@ class Inference:
         if api is None:
             api = self.api
         return api
-
+        
     def _inference_auto(
         self,
         source: List[Union[str, np.ndarray]],
         settings: Dict[str, Any],
     ) -> Tuple[List[Annotation], List[dict]]:
+        self._unfreeze_model()
         inference_mode = settings.get("inference_mode", "full_image")
         use_raw = (
             inference_mode == "sliding_window" and settings["sliding_window_mode"] == "advanced"
@@ -1453,9 +1464,11 @@ class Inference:
         if (not use_raw and self.is_batch_inference_supported()) or (
             use_raw and is_predict_batch_raw_implemented
         ):
-            return self._inference_batched_wrapper(source, settings)
+            result = self._inference_batched_wrapper(source, settings)
         else:
-            return self._inference_one_by_one_wrapper(source, settings)
+            result = self._inference_one_by_one_wrapper(source, settings)
+        self._schedule_freeze_on_inactivity()
+        return result
 
     def inference(
         self,
@@ -2432,9 +2445,7 @@ class Inference:
     def _check_serve_before_call(self, func):
         @wraps(func)
         def wrapper(*args, **kwargs):
-            if self._model_served is True:
-                return func(*args, **kwargs)
-            else:
+            if self._model_served is False:
                 msg = (
                     "The model has not yet been deployed. "
                     "Please select the appropriate model in the UI and press the 'Serve' button. "
@@ -2442,9 +2453,53 @@ class Inference:
                 )
                 # raise DialogWindowError(title="Call undeployed model.", description=msg)
                 raise RuntimeError(msg)
-
+            return func(*args, **kwargs)
         return wrapper
 
+    def _freeze_model(self):
+        if self._model_frozen or not self._model_served:
+            return
+        logger.debug("Freezing model...")
+        runtime = self._deploy_params.get("runtime")
+        if runtime and runtime.lower() != RuntimeType.PYTORCH.lower():
+            logger.debug("Model is not running in PyTorch runtime, cannot freeze.")
+            return
+        previous_device = self._deploy_params.get("device")
+        if previous_device == "cpu":
+            logger.debug("Model is already running on CPU, cannot freeze.")
+            return
+
+        deploy_params = self._deploy_params.copy()
+        deploy_params["device"] = "cpu"
+        try:
+            self._load_model(deploy_params)
+            self._model_frozen = True
+            logger.info(
+                "Model has been re-deployed to CPU for resource optimization. "
+                "It will be loaded back to the original device on the next inference request."
+            )
+        finally:
+            self._deploy_params["device"] = previous_device
+            clean_up_cuda()
+
+    def _unfreeze_model(self):
+        if not self._model_frozen:
+            return
+        logger.debug("Unfreezing model...")
+        self._model_frozen = False
+        self._load_model(self._deploy_params)
+        clean_up_cuda()
+        logger.debug("Model is unfrozen and ready for inference.")
+
+    def _schedule_freeze_on_inactivity(self):
+        if self._freeze_timer is not None:
+            self._freeze_timer.cancel()
+        timer = threading.Timer(self._inactivity_timeout, self._freeze_model)
+        timer.daemon = True
+        timer.start()
+        self._freeze_timer = timer
+        logger.debug("Model will be frozen in %s seconds due to inactivity.", self._inactivity_timeout)
+
     def _set_served_callback(self):
         self._model_served = True
 
@@ -2506,6 +2561,7 @@ class Inference:
             # update to set correct device
             device = deploy_params.get("device", "cpu")
             self.gui.set_deployed(device)
+            self._schedule_freeze_on_inactivity()
             return {"result": "model was successfully deployed"}
         except Exception as e:
             self.gui._success_label.hide()
@@ -3400,6 +3456,8 @@ class Inference:
                 if self.gui is not None:
                     self.gui._success_label.hide()
                 raise e
+            finally:
+                self._schedule_freeze_on_inactivity()
 
         @server.post("/list_pretrained_models")
         def _list_pretrained_models():
@@ -3479,6 +3537,16 @@ class Inference:
                 },
             }
 
+        @server.post("/freeze_model")
+        def _freeze_model(request: Request):
+            if self._model_frozen:
+                return {"message": "Model is already frozen."}
+
+            self._freeze_model()
+            if not self._model_frozen:
+                return {"message": "Failed to freeze model. Check the logs for details."}
+            return {"message": "Model is frozen."}
+
         # Local deploy without predict args
         if self._is_cli_deploy:
             self._run_server()

supervisely/nn/model/model_api.py

@@ -111,6 +111,12 @@ class ModelAPI:
         response = self._post("tasks.stop", {ApiField.ID: id})
         return TaskApi.Status(response[ApiField.STATUS])
 
+    def freeze_model(self):
+        """Freeze the model to free up resources."""
+        if self.task_id is not None:
+            return self.api.task.send_request(self.task_id, "freeze_model", {})
+        return self._post("freeze_model", {})
+
     # --------------------- #
 
     # region Load

{supervisely-6.73.406.dist-info → supervisely-6.73.408.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: supervisely
-Version: 6.73.406
+Version: 6.73.408
 Summary: Supervisely Python SDK.
 Home-page: https://github.com/supervisely/supervisely
 Author: Supervisely

{supervisely-6.73.406.dist-info → supervisely-6.73.408.dist-info}/RECORD

@@ -23,7 +23,7 @@ supervisely/api/advanced_api.py,sha256=Nd5cCnHFWc3PSUrCtENxTGtDjS37_lCHXsgXvUI3T
 supervisely/api/agent_api.py,sha256=8EQBwD6v7KLS0-xKcZ12B7mtzKwG7RRgq1fk1vaN144,8893
 supervisely/api/annotation_api.py,sha256=U6dHUIOt6Fe8XcbX1MA19z-fg91maOumJAawKG5ZJsk,82876
 supervisely/api/api.py,sha256=pEgRIWlVqDdtDjAeL_nx2Rwldm6ANwLLacm6kLnyvbE,67723
-supervisely/api/app_api.py,sha256=HaltkadXqel7hFBQdwsn5Qvmrx5eEqDXUu6X98nHerk,76226
+supervisely/api/app_api.py,sha256=ghhBVaxca7xO9Bgy2zbexC89BJ9s0k7fVPd6-6F_Wvw,76594
 supervisely/api/constants.py,sha256=WfqIcEpRnU4Mcfb6q0njeRs2VVSoTAJaIyrqBkBjP8I,253
 supervisely/api/dataset_api.py,sha256=7idBMFL8jumWNw-wlBAbQWC09RskG-3GlidfPDukq3Q,47930
 supervisely/api/entities_collection_api.py,sha256=Be13HsfMFLmq9XpiOfQog0Y569kbUn52hXv6x5vX3Vg,22624
@@ -44,7 +44,7 @@ supervisely/api/remote_storage_api.py,sha256=1O4rTIwW8s9gxC00yvFuKbEMGNsa7YSRlZ8
 supervisely/api/report_api.py,sha256=Om7CGulUbQ4BuJ16eDtz7luLe0JQNqab-LoLpUXu7YE,7123
 supervisely/api/role_api.py,sha256=c1XAU_wZg6zL4wG2R7iuS9EJOoaHHNGchxa1nYVL7yo,3047
 supervisely/api/storage_api.py,sha256=VxiflQt-SfyB1OuEOB66JsMkxCosUr4n0WHQ5if3Ltg,13039
-supervisely/api/task_api.py,sha256=eRDeUj9vexzJl_vQqW-DDYUkHaWfFqNg_EQI597RE0Y,36678
+supervisely/api/task_api.py,sha256=5Ae3jKWgvvz_Mljk6V1UaIVqDbTvnSD6cpIJ3H3KqT0,38435
 supervisely/api/team_api.py,sha256=uPsBpDP_Ig9HDQ9Zm6Y-VboLbSYKIV9S_a1S7e4vqvo,19470
 supervisely/api/user_api.py,sha256=m29GP9tvem8P2fJZgg7DAZ9yhFdBX26ZBcWxCKdnhn4,24943
 supervisely/api/video_annotation_tool_api.py,sha256=3A9-U8WJzrTShP_n9T8U01M9FzGYdeS51CCBTzUnooo,6686
@@ -893,7 +893,7 @@ supervisely/nn/benchmark/visualization/widgets/table/__init__.py,sha256=47DEQpj8
 supervisely/nn/benchmark/visualization/widgets/table/table.py,sha256=atmDnF1Af6qLQBUjLhK18RMDKAYlxnsuVHMSEa5a-e8,4319
 supervisely/nn/inference/__init__.py,sha256=QFukX2ip-U7263aEPCF_UCFwj6EujbMnsgrXp5Bbt8I,1623
 supervisely/nn/inference/cache.py,sha256=rfmb1teJ9lNDfisUSh6bwDCVkPZocn8GMvDgLQktnbo,35023
-supervisely/nn/inference/inference.py,sha256=L14M8qGofz6EGrgJvGMcz0h3vsu-AnHPeL_iUO8Kf3Y,195721
+supervisely/nn/inference/inference.py,sha256=7c2-GuG3MgI5H0muTyxoR-XgUzWpbP9if37CyRewobA,198528
 supervisely/nn/inference/inference_request.py,sha256=y6yw0vbaRRcEBS27nq3y0sL6Gmq2qLA_Bm0GrnJGegE,14267
 supervisely/nn/inference/session.py,sha256=dIg2F-OBl68pUzcmtmcI0YQIp1WWNnrJTVMjwFN91Q4,35824
 supervisely/nn/inference/uploader.py,sha256=21a9coOimCHhEqAbV-llZWcp12847DEMoQp3N16bpK0,5425
@@ -968,7 +968,7 @@ supervisely/nn/legacy/pytorch/weights.py,sha256=Zb9kcpUCg6ykr7seO53CkKSQa2K44wo8
 supervisely/nn/legacy/training/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 supervisely/nn/legacy/training/eval_planner.py,sha256=zN9b0_CX7sWGdC8e6riTvD-NOUc3_Xduyhj00S7PEIo,1311
 supervisely/nn/model/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-supervisely/nn/model/model_api.py,sha256=rIne6ymvL2aVwzSTJMGxAbevE0U-f-WKhDciqaneqRc,9666
+supervisely/nn/model/model_api.py,sha256=rq-08BrQmWKqUxWnBB5yKBfnjxw8Lg88mhva848Ae2I,9911
 supervisely/nn/model/prediction.py,sha256=N3oO9s3NDiC5CFvW8utfU8rz3bfpCl37Sk4VEBH94Bc,11307
 supervisely/nn/model/prediction_session.py,sha256=sy0FSQaWSmT8i0RkR4J8oIn3Ek4IDVJNBR1Tg4mulkM,25523
 supervisely/nn/tracker/__init__.py,sha256=LiojByb5kGsTQ49lWuboEh7B4JUwM1vfz81J8kJlLYo,337
@@ -1114,9 +1114,9 @@ supervisely/worker_proto/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZ
 supervisely/worker_proto/worker_api_pb2.py,sha256=VQfi5JRBHs2pFCK1snec3JECgGnua3Xjqw_-b3aFxuM,59142
 supervisely/worker_proto/worker_api_pb2_grpc.py,sha256=3BwQXOaP9qpdi0Dt9EKG--Lm8KGN0C5AgmUfRv77_Jk,28940
 supervisely_lib/__init__.py,sha256=7-3QnN8Zf0wj8NCr2oJmqoQWMKKPKTECvjH9pd2S5vY,159
-supervisely-6.73.406.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-supervisely-6.73.406.dist-info/METADATA,sha256=696rYpRYkDoHehnuB_Y2ngM5dHHY_hX0qlMyiyZ9PqA,35254
-supervisely-6.73.406.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-supervisely-6.73.406.dist-info/entry_points.txt,sha256=U96-5Hxrp2ApRjnCoUiUhWMqijqh8zLR03sEhWtAcms,102
-supervisely-6.73.406.dist-info/top_level.txt,sha256=kcFVwb7SXtfqZifrZaSE3owHExX4gcNYe7Q2uoby084,28
-supervisely-6.73.406.dist-info/RECORD,,
+supervisely-6.73.408.dist-info/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+supervisely-6.73.408.dist-info/METADATA,sha256=C6yRG5xZiYpAXpe3RyAIimpbPRggSn--0fqKT00sSKI,35254
+supervisely-6.73.408.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+supervisely-6.73.408.dist-info/entry_points.txt,sha256=U96-5Hxrp2ApRjnCoUiUhWMqijqh8zLR03sEhWtAcms,102
+supervisely-6.73.408.dist-info/top_level.txt,sha256=kcFVwb7SXtfqZifrZaSE3owHExX4gcNYe7Q2uoby084,28
+supervisely-6.73.408.dist-info/RECORD,,