mlrun 1.10.0rc11__tar.gz → 1.10.0rc13__tar.gz

{mlrun-1.10.0rc11/mlrun.egg-info → mlrun-1.10.0rc13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mlrun
-Version: 1.10.0rc11
+Version: 1.10.0rc13
 Summary: Tracking and config of machine learning runs
 Home-page: https://github.com/mlrun/mlrun
 Author: Yaron Haviv
@@ -31,7 +31,7 @@ Requires-Dist: ipython~=8.10
 Requires-Dist: nuclio-jupyter~=0.11.1
 Requires-Dist: numpy<1.27.0,>=1.26.4
 Requires-Dist: pandas<2.2,>=1.2
-Requires-Dist: pyarrow<17,>=10.0
+Requires-Dist: pyarrow<18,>=10.0
 Requires-Dist: pyyaml<7,>=6.0.2
 Requires-Dist: requests~=2.32
 Requires-Dist: tabulate~=0.8.6

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/__init__.py

@@ -32,7 +32,7 @@ from typing import Optional
 import dotenv
 
 from .config import config as mlconf
-from .datastore import DataItem, store_manager
+from .datastore import DataItem, ModelProvider, store_manager
 from .db import get_run_db
 from .errors import MLRunInvalidArgumentError, MLRunNotFoundError
 from .execution import MLClientCtx
@@ -55,6 +55,7 @@ from .run import (
     code_to_function,
     function_to_module,
     get_dataitem,
+    get_model_provider,
     get_object,
     get_or_create_ctx,
     get_pipeline,

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/__main__.py

@@ -261,7 +261,13 @@ def run(
     config = environ.get("MLRUN_EXEC_CONFIG")
     if from_env and config:
         config = json.loads(config)
-        runobj = RunTemplate.from_dict(config)
+        # If run is a retry we need to maintain the run status therefore using RunObject instead of RunTemplate
+        retry_count = config.get("status", {}).get("retry_count")
+        if retry_count:
+            logger.info(f"Retrying run - attempt: {retry_count + 1}")
+            runobj = mlrun.RunObject.from_dict(config)
+        else:
+            runobj = RunTemplate.from_dict(config)
     elif task:
         obj = get_object(task)
         task = yaml.load(obj, Loader=yaml.FullLoader)

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/artifacts/base.py

@@ -839,9 +839,7 @@ def get_artifact_meta(artifact):
         artifact = artifact.artifact_url
 
     if mlrun.datastore.is_store_uri(artifact):
-        artifact_spec, target = mlrun.datastore.store_manager.get_store_artifact(
-            artifact
-        )
+        artifact_spec, _ = mlrun.datastore.store_manager.get_store_artifact(artifact)
 
     elif artifact.lower().endswith(".yaml"):
         data = mlrun.datastore.store_manager.object(url=artifact).get()
@@ -942,3 +940,11 @@ def fill_artifact_object_hash(object_dict, iteration=None, producer_id=None):
             object_dict["spec"][key] = value
 
     return uid
+
+
+def verify_target_path(artifact: Artifact):
+    if not artifact.get_target_path():
+        raise mlrun.errors.MLRunInvalidArgumentError(
+            f"artifact {artifact.uri} "
+            f"does not have a valid/persistent offline target"
+        )

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/artifacts/dataset.py

@@ -26,7 +26,7 @@ import mlrun.datastore
 import mlrun.utils.helpers
 from mlrun.config import config as mlconf
 
-from .base import Artifact, ArtifactSpec, StorePrefix
+from .base import Artifact, ArtifactSpec, StorePrefix, verify_target_path
 
 default_preview_rows_length = 20
 max_preview_columns = mlconf.artifacts.datasets.max_preview_columns
@@ -424,6 +424,7 @@ def update_dataset_meta(
         artifact_spec = artifact
     elif mlrun.datastore.is_store_uri(artifact):
         artifact_spec, _ = mlrun.datastore.store_manager.get_store_artifact(artifact)
+        verify_target_path(artifact_spec)
     else:
         raise ValueError("model path must be a model store object/URL/DataItem")
 

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/artifacts/llm_prompt.py

@@ -61,7 +61,11 @@ class LLMPromptArtifactSpec(ArtifactSpec):
         self.prompt_legend = prompt_legend
         self.model_configuration = model_configuration
         self.description = description
-        self._model_artifact = None
+        self._model_artifact = (
+            model_artifact
+            if isinstance(model_artifact, model_art.ModelArtifact)
+            else None
+        )
 
     @property
     def model_uri(self):
@@ -127,7 +131,7 @@ class LLMPromptArtifact(Artifact):
         if self.spec._model_artifact:
             return self.spec._model_artifact
         if self.spec.model_uri:
-            self.spec._model_artifact, target = (
+            self.spec._model_artifact, _ = (
                 mlrun.datastore.store_manager.get_store_artifact(self.spec.model_uri)
             )
             return self.spec._model_artifact

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/artifacts/model.py

@@ -26,7 +26,7 @@ from ..data_types import InferOptions, get_infer_interface
 from ..features import Feature
 from ..model import ObjectList
 from ..utils import StorePrefix, is_relative_path
-from .base import Artifact, ArtifactSpec, upload_extra_data
+from .base import Artifact, ArtifactSpec, upload_extra_data, verify_target_path
 
 model_spec_filename = "model_spec.yaml"
 MODEL_OPTIONAL_SUFFIXES = [".tar.gz", ".pkl", ".bin", ".pickle"]
@@ -493,7 +493,6 @@ def get_model(
     :returns: model filename, model artifact object, extra data dict
 
     """
-    # TODO support LLMPromptArtifact
     model_file = ""
     model_spec = None
     extra_dataitems = {}
@@ -518,6 +517,7 @@ def get_model(
             model_spec, target = mlrun.datastore.store_manager.get_store_artifact(
                 model_dir
             )
+            verify_target_path(model_spec)
         else:
             model_spec, target = model_dir, model_dir.get_target_path()
         if not model_spec or model_spec.kind != "model":

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/common/constants.py

@@ -66,6 +66,7 @@ class MLRunInternalLabels:
     scrape_metrics = f"{MLRUN_LABEL_PREFIX}scrape-metrics"
     tag = f"{MLRUN_LABEL_PREFIX}tag"
     uid = f"{MLRUN_LABEL_PREFIX}uid"
+    retry = f"{MLRUN_LABEL_PREFIX}retry-attempt"
     username = f"{MLRUN_LABEL_PREFIX}username"
     username_domain = f"{MLRUN_LABEL_PREFIX}username_domain"
     task_name = f"{MLRUN_LABEL_PREFIX}task-name"

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/common/runtimes/constants.py

@@ -139,6 +139,7 @@ class RunStates:
     aborted = "aborted"
     aborting = "aborting"
     skipped = "skipped"
+    pending_retry = "pendingRetry"
 
     @staticmethod
     def all():
@@ -152,6 +153,7 @@ class RunStates:
             RunStates.aborted,
             RunStates.aborting,
             RunStates.skipped,
+            RunStates.pending_retry,
         ]
 
     @staticmethod
@@ -168,6 +170,7 @@ class RunStates:
         return [
             RunStates.error,
             RunStates.aborted,
+            RunStates.pending_retry,
         ]
 
     @staticmethod
@@ -185,12 +188,18 @@ class RunStates:
     def non_terminal_states():
         return list(set(RunStates.all()) - set(RunStates.terminal_states()))
 
+    @staticmethod
+    def terminal_or_error_states():
+        return list(
+            set(RunStates.terminal_states())
+            | set(RunStates.error_and_abortion_states())
+        )
+
     @staticmethod
     def not_allowed_for_deletion_states():
         return [
             RunStates.running,
             RunStates.pending,
-            # TODO: add aborting state once we have it
         ]
 
     @staticmethod

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/common/schemas/__init__.py

@@ -214,7 +214,7 @@ from .secret import (
     SecretsData,
     UserSecretCreationRequest,
 )
-from .serving import ModelRunnerStepData, MonitoringData
+from .serving import ModelRunnerStepData, ModelsData, MonitoringData
 from .tag import Tag, TagObjects
 from .workflow import (
     GetWorkflowResponse,

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/common/schemas/model_monitoring/model_endpoints.py

@@ -336,8 +336,8 @@ class ModelEndpointMonitoringMetricNoData(_ModelEndpointMonitoringMetricValuesBa
 
 class ApplicationBaseRecord(BaseModel):
     type: Literal["metric", "result"]
-    time: datetime
     value: float
+    time: Optional[datetime] = None
 
 
 class ApplicationResultRecord(ApplicationBaseRecord):

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/common/schemas/serving.py

@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import enum
+
 from pydantic.v1 import BaseModel
 
 from mlrun.common.types import StrEnum
@@ -40,3 +42,8 @@ class MonitoringData(StrEnum):
     MODEL_PATH = "model_path"
     MODEL_ENDPOINT_UID = "model_endpoint_uid"
     MODEL_CLASS = "model_class"
+
+
+class ModelsData(enum.Enum):
+    MODEL_CLASS = 0
+    MODEL_PARAMETERS = 1

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/config.py

@@ -120,6 +120,14 @@ default_config = {
             # max number of parallel abort run jobs in runs monitoring
             "concurrent_abort_stale_runs_workers": 10,
             "list_runs_time_period_in_days": 7,  # days
+            "retry": {
+                # periodic job for triggering retries interval in seconds
+                "interval": "30",
+                # runs limit to fetch for retrying
+                "fetch_runs_limit": 1000,
+                # minutes until a run is considered stale and will be aborted
+                "staleness_threshold": 60 * 24 * 3,
+            },
         },
         "projects": {
             "summaries": {
@@ -184,6 +192,9 @@ default_config = {
         "url": "",
     },
     "v3io_framesd": "http://framesd:8080",
+    "model_providers": {
+        "openai_default_model": "gpt-4",
+    },
     # default node selector to be applied to all functions - json string base64 encoded format
     "default_function_node_selector": "e30=",
     # default priority class to be applied to functions running on k8s cluster
@@ -270,6 +281,12 @@ default_config = {
                     "executing": "24h",
                 }
             },
+            "retry": {
+                "backoff": {
+                    "default_base_delay": "30s",
+                    "min_base_delay": "30s",
+                },
+            },
             # When the module is reloaded, the maximum depth recursion configuration for the recursive reload
             # function is used to prevent infinite loop
             "reload_max_recursion_depth": 100,
@@ -316,6 +333,7 @@ default_config = {
                     "project_summaries": "enabled",
                     "start_logs": "enabled",
                     "stop_logs": "enabled",
+                    "retry_jobs": "enabled",
                 },
             },
             "worker": {
@@ -539,7 +557,7 @@ default_config = {
         },
         "v3io_api": "",
         "v3io_framesd": "",
-        # If running from sdk and MLRUN_DBPATH is not set, the db will fallback to a nop db which will not preform any
+        # If running from sdk and MLRUN_DBPATH is not set, the db will fallback to a nop db which will not perform any
         # run db operations.
         "nop_db": {
             # if set to true, will raise an error for trying to use run db functionality
@@ -641,7 +659,7 @@ default_config = {
         "offline_storage_path": "model-endpoints/{kind}",
         "parquet_batching_max_events": 10_000,
         "parquet_batching_timeout_secs": timedelta(minutes=1).total_seconds(),
-        "model_endpoint_creation_check_period": "15",
+        "model_endpoint_creation_check_period": 15,
     },
     "secret_stores": {
         # Use only in testing scenarios (such as integration tests) to avoid using k8s for secrets (will use in-memory
@@ -1219,6 +1237,7 @@ class Config:
         """
         Get the default value for the ssl_redirect configuration.
         In Iguazio we always want to redirect to HTTPS, in other cases we don't.
+
         :return: True if we should redirect to HTTPS, False otherwise.
         """
         return self.is_running_on_iguazio()

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/datastore/__init__.py

@@ -14,6 +14,7 @@
 
 __all__ = [
     "DataItem",
+    "ModelProvider",
     "get_store_resource",
     "ParquetTarget",
     "CSVTarget",
@@ -32,12 +33,12 @@ __all__ = [
     "get_stream_pusher",
     "ConfigProfile",
     "VectorStoreCollection",
+    "store_manager",
 ]
 
 from urllib.parse import urlparse
 
 import fsspec
-from mergedeep import merge
 
 import mlrun.datastore.wasbfs
 from mlrun.datastore.datastore_profile import (
@@ -45,6 +46,7 @@ from mlrun.datastore.datastore_profile import (
     DatastoreProfileKafkaTarget,
     DatastoreProfileV3io,
 )
+from mlrun.datastore.model_provider.model_provider import ModelProvider
 from mlrun.platforms.iguazio import (
     HTTPOutputStream,
     KafkaOutputStream,

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/datastore/alibaba_oss.py

@@ -69,7 +69,7 @@ class OSSStore(DataStore):
             key=self._get_secret_or_env("ALIBABA_ACCESS_KEY_ID"),
             secret=self._get_secret_or_env("ALIBABA_SECRET_ACCESS_KEY"),
         )
-        return self._sanitize_storage_options(res)
+        return self._sanitize_options(res)
 
     def get_bucket_and_key(self, key):
         path = self._join(key)[1:]

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/datastore/azure_blob.py

@@ -67,7 +67,7 @@ class AzureBlobStore(DataStore):
                 or self._get_secret_or_env("AZURE_STORAGE_SAS_TOKEN"),
                 credential=self._get_secret_or_env("credential"),
             )
-            self._storage_options = self._sanitize_storage_options(res)
+            self._storage_options = self._sanitize_options(res)
         return self._storage_options
 
     @property

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/datastore/base.py

@@ -28,6 +28,7 @@ import requests
 
 import mlrun.config
 import mlrun.errors
+from mlrun.datastore.remote_client import BaseRemoteClient
 from mlrun.errors import err_to_str
 from mlrun.utils import StorePrefix, is_jupyter, logger
 
@@ -45,22 +46,19 @@ class FileStats:
         return f"FileStats(size={self.size}, modified={self.modified}, type={self.content_type})"
 
 
-class DataStore:
+class DataStore(BaseRemoteClient):
     using_bucket = False
 
     def __init__(
         self, parent, name, kind, endpoint="", secrets: Optional[dict] = None, **kwargs
     ):
-        self._parent = parent
-        self.kind = kind
-        self.name = name
-        self.endpoint = endpoint
+        super().__init__(
+            parent=parent, kind=kind, name=name, endpoint=endpoint, secrets=secrets
+        )
         self.subpath = ""
-        self.secret_pfx = ""
         self.options = {}
         self.from_spec = False
         self._filesystem = None
-        self._secrets = secrets or {}
 
     @property
     def is_structured(self):
@@ -70,13 +68,6 @@ class DataStore:
     def is_unstructured(self):
         return True
 
-    @staticmethod
-    def _sanitize_storage_options(options):
-        if not options:
-            return {}
-        options = {k: v for k, v in options.items() if v is not None and v != ""}
-        return options
-
     @staticmethod
     def _sanitize_url(url):
         """
@@ -106,15 +97,9 @@ class DataStore:
         """Whether the data store supports isdir"""
         return True
 
-    def _get_secret_or_env(self, key, default=None):
-        # Project-secrets are mounted as env variables whose name can be retrieved from SecretsStore
-        return mlrun.get_secret_or_env(
-            key, secret_provider=self._get_secret, default=default
-        )
-
     def get_storage_options(self):
         """get fsspec storage options"""
-        return self._sanitize_storage_options(None)
+        return self._sanitize_options(None)
 
     def open(self, filepath, mode):
         file_system = self.filesystem
@@ -125,16 +110,6 @@ class DataStore:
             return f"{self.subpath}/{key}"
         return key
 
-    def _get_parent_secret(self, key):
-        return self._parent.secret(self.secret_pfx + key)
-
-    def _get_secret(self, key: str, default=None):
-        return self._secrets.get(key, default) or self._get_parent_secret(key)
-
-    @property
-    def url(self):
-        return f"{self.kind}://{self.endpoint}"
-
     @property
     def spark_url(self):
         return self.url

{mlrun-1.10.0rc11 → mlrun-1.10.0rc13}/mlrun/datastore/datastore.py

@@ -11,50 +11,40 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 import warnings
+from functools import partial
 from typing import Optional
-from urllib.parse import urlparse
 
 from mergedeep import merge
 
 import mlrun
 import mlrun.errors
+from mlrun.artifacts.llm_prompt import LLMPromptArtifact
+from mlrun.artifacts.model import ModelArtifact
 from mlrun.datastore.datastore_profile import datastore_profile_read
+from mlrun.datastore.model_provider.model_provider import (
+    ModelProvider,
+)
+from mlrun.datastore.remote_client import BaseRemoteClient
+from mlrun.datastore.utils import (
+    parse_url,
+)
 from mlrun.errors import err_to_str
 from mlrun.utils.helpers import get_local_file_schema
 
+from ..artifacts.base import verify_target_path
 from ..utils import DB_SCHEMA, RunKeys
 from .base import DataItem, DataStore, HttpStore
 from .filestore import FileStore
 from .inmem import InMemoryStore
+from .model_provider.openai_provider import OpenAIProvider
 from .store_resources import get_store_resource, is_store_uri
 from .v3io import V3ioStore
 
 in_memory_store = InMemoryStore()
 
 
-def parse_url(url):
-    if url and url.startswith("v3io://") and not url.startswith("v3io:///"):
-        url = url.replace("v3io://", "v3io:///", 1)
-    parsed_url = urlparse(url)
-    schema = parsed_url.scheme.lower()
-    endpoint = parsed_url.hostname
-    if endpoint:
-        # HACK - urlparse returns the hostname after in lower case - we want the original case:
-        # the hostname is a substring of the netloc, in which it's the original case, so we find the indexes of the
-        # hostname in the netloc and take it from there
-        lower_hostname = parsed_url.hostname
-        netloc = str(parsed_url.netloc)
-        lower_netloc = netloc.lower()
-        hostname_index_in_netloc = lower_netloc.index(str(lower_hostname))
-        endpoint = netloc[
-            hostname_index_in_netloc : hostname_index_in_netloc + len(lower_hostname)
-        ]
-    if parsed_url.port:
-        endpoint += f":{parsed_url.port}"
-    return schema, endpoint, parsed_url
-
-
 def schema_to_store(schema) -> DataStore.__subclasses__():
     # import store classes inside to enable making their dependencies optional (package extras)
 
@@ -109,6 +99,20 @@ def schema_to_store(schema) -> DataStore.__subclasses__():
     raise ValueError(f"unsupported store scheme ({schema})")
 
 
+def schema_to_model_provider(
+    schema: str, raise_missing_schema_exception=True
+) -> type[ModelProvider]:
+    #  TODO add hugging face and http
+    schema_dict = {"openai": OpenAIProvider}
+    provider_class = schema_dict.get(schema, None)
+    if not provider_class:
+        if raise_missing_schema_exception:
+            raise ValueError(f"unsupported model provider schema ({schema})")
+        else:
+            warnings.warn(f"unsupported model provider schema: {schema}")
+    return provider_class
+
+
 def uri_to_ipython(link):
     schema, endpoint, parsed_url = parse_url(link)
     if schema in [DB_SCHEMA, "memory", "ds"]:
@@ -159,7 +163,11 @@ class StoreManager:
         self._stores[store.name] = store
 
     def get_store_artifact(
-        self, url, project="", allow_empty_resources=None, secrets=None
+        self,
+        url,
+        project="",
+        allow_empty_resources=None,
+        secrets=None,
     ):
         """
         This is expected to be run only on client side. server is not expected to load artifacts.
@@ -175,12 +183,21 @@ class StoreManager:
         except Exception as exc:
             raise OSError(f"artifact {url} not found, {err_to_str(exc)}")
         target = resource.get_target_path()
+
         # the allow_empty.. flag allows us to have functions which dont depend on having targets e.g. a function
         # which accepts a feature vector uri and generate the offline vector (parquet) for it if it doesnt exist
-        if not target and not allow_empty_resources:
-            raise mlrun.errors.MLRunInvalidArgumentError(
-                f"Resource {url} does not have a valid/persistent offline target"
-            )
+        if not allow_empty_resources:
+            if isinstance(resource, LLMPromptArtifact):
+                if not resource.spec.model_uri:
+                    raise mlrun.errors.MLRunInvalidArgumentError(
+                        f"LLMPromptArtifact {url} does not contain model artifact uri"
+                    )
+            elif not target and not (
+                isinstance(resource, ModelArtifact) and resource.model_url
+            ):
+                raise mlrun.errors.MLRunInvalidArgumentError(
+                    f"Resource {url} does not have a valid/persistent offline target or model_url"
+                )
         return resource, target or ""
 
     def object(
@@ -190,6 +207,7 @@ class StoreManager:
         project="",
         allow_empty_resources=None,
         secrets: Optional[dict] = None,
+        **kwargs,
     ) -> DataItem:
         meta = artifact_url = None
         if is_store_uri(url):
@@ -197,6 +215,8 @@ class StoreManager:
             meta, url = self.get_store_artifact(
                 url, project, allow_empty_resources, secrets
             )
+            if not allow_empty_resources:
+                verify_target_path(meta)
 
         store, subpath, url = self.get_or_create_store(
             url, secrets=secrets, project_name=project
@@ -218,7 +238,7 @@ class StoreManager:
         cache: Optional[dict] = None,
         schema_to_class: callable = schema_to_store,
         **kwargs,
-    ) -> (DataStore, str, str):
+    ) -> (BaseRemoteClient, str, str):
         # The cache can be an empty dictionary ({}), even if it is a _stores object
         cache = cache if cache is not None else {}
         schema, endpoint, parsed_url = parse_url(url)
@@ -227,10 +247,7 @@ class StoreManager:
 
         if schema == "ds":
             datastore_profile = datastore_profile_read(url, project_name, secrets)
-            if secrets and datastore_profile.secrets():
-                secrets = merge(secrets, datastore_profile.secrets())
-            else:
-                secrets = secrets or datastore_profile.secrets()
+            secrets = merge(secrets or {}, datastore_profile.secrets() or {})
             url = datastore_profile.url(subpath)
             schema, endpoint, parsed_url = parse_url(url)
             subpath = parsed_url.path
@@ -260,6 +277,9 @@ class StoreManager:
         remote_client_class = schema_to_class(schema)
         remote_client = None
         if remote_client_class:
+            endpoint, subpath = remote_client_class.parse_endpoint_and_path(
+                endpoint, subpath
+            )
             remote_client = remote_client_class(
                 self, schema, cache_key, parsed_url.netloc, secrets=secrets, **kwargs
             )
@@ -288,5 +308,61 @@ class StoreManager:
             )
         return datastore, sub_path, url
 
+    def get_or_create_model_provider(
+        self,
+        url,
+        secrets: Optional[dict] = None,
+        project_name="",
+        default_invoke_kwargs: Optional[dict] = None,
+        raise_missing_schema_exception=True,
+    ) -> ModelProvider:
+        schema_to_provider_with_raise = partial(
+            schema_to_model_provider,
+            raise_missing_schema_exception=raise_missing_schema_exception,
+        )
+        model_provider, _, _ = self._get_or_create_remote_client(
+            url=url,
+            secrets=secrets,
+            project_name=project_name,
+            schema_to_class=schema_to_provider_with_raise,
+            default_invoke_kwargs=default_invoke_kwargs,
+        )
+        if model_provider and not isinstance(model_provider, ModelProvider):
+            raise mlrun.errors.MLRunInvalidArgumentError(
+                "remote client by url is not model_provider"
+            )
+        return model_provider
+
     def reset_secrets(self):
         self._secrets = {}
+
+    def model_provider_object(
+        self,
+        url,
+        project="",
+        allow_empty_resources=None,
+        secrets: Optional[dict] = None,
+        default_invoke_kwargs: Optional[dict] = None,
+        raise_missing_schema_exception=True,
+    ) -> ModelProvider:
+        if mlrun.datastore.is_store_uri(url):
+            resource = self.get_store_artifact(
+                url,
+                project,
+                allow_empty_resources,
+                secrets,
+            )
+            if not isinstance(resource, ModelArtifact) or not resource.model_url:
+                raise mlrun.errors.MLRunInvalidArgumentError(
+                    "unable to create the model provider from the given resource URI"
+                )
+            url = resource.model_url
+            default_invoke_kwargs = default_invoke_kwargs or resource.default_config
+        model_provider = self.get_or_create_model_provider(
+            url,
+            secrets=secrets,
+            project_name=project,
+            default_invoke_kwargs=default_invoke_kwargs,
+            raise_missing_schema_exception=raise_missing_schema_exception,
+        )
+        return model_provider