oracle-ads 2.12.2__py3-none-any.whl → 2.12.3__py3-none-any.whl

ads/aqua/common/enums.py

@@ -52,6 +52,7 @@ class InferenceContainerTypeFamily(str, metaclass=ExtendedEnumMeta):
     AQUA_VLLM_CONTAINER_FAMILY = "odsc-vllm-serving"
     AQUA_TGI_CONTAINER_FAMILY = "odsc-tgi-serving"
     AQUA_LLAMA_CPP_CONTAINER_FAMILY = "odsc-llama-cpp-serving"
+    AQUA_TEI_CONTAINER_FAMILY = "odsc-tei-serving"
 
 
 class InferenceContainerParamType(str, metaclass=ExtendedEnumMeta):
@@ -80,3 +81,11 @@ class RqsAdditionalDetails(str, metaclass=ExtendedEnumMeta):
     MODEL_VERSION_SET_NAME = "modelVersionSetName"
     PROJECT_ID = "projectId"
     VERSION_LABEL = "versionLabel"
+
+
+class TextEmbeddingInferenceContainerParams(str, metaclass=ExtendedEnumMeta):
+    """Contains a subset of params that are required for enabling model deployment in OCI Data Science. More options
+    are available at https://huggingface.co/docs/text-embeddings-inference/en/cli_arguments"""
+
+    MODEL_ID = "model-id"
+    PORT = "port"

ads/aqua/common/utils.py

@@ -35,6 +35,7 @@ from ads.aqua.common.enums import (
     InferenceContainerParamType,
     InferenceContainerType,
     RqsAdditionalDetails,
+    TextEmbeddingInferenceContainerParams,
 )
 from ads.aqua.common.errors import (
     AquaFileNotFoundError,
@@ -51,6 +52,7 @@ from ads.aqua.constants import (
     MODEL_BY_REFERENCE_OSS_PATH_KEY,
     SERVICE_MANAGED_CONTAINER_URI_SCHEME,
     SUPPORTED_FILE_FORMATS,
+    TEI_CONTAINER_DEFAULT_HOST,
     TGI_INFERENCE_RESTRICTED_PARAMS,
     UNKNOWN,
     UNKNOWN_JSON_STR,
@@ -63,7 +65,12 @@ from ads.common.extended_enum import ExtendedEnumMeta
 from ads.common.object_storage_details import ObjectStorageDetails
 from ads.common.oci_resource import SEARCH_TYPE, OCIResource
 from ads.common.utils import copy_file, get_console_link, upload_to_os
-from ads.config import AQUA_SERVICE_MODELS_BUCKET, CONDA_BUCKET_NS, TENANCY_OCID
+from ads.config import (
+    AQUA_MODEL_DEPLOYMENT_FOLDER,
+    AQUA_SERVICE_MODELS_BUCKET,
+    CONDA_BUCKET_NS,
+    TENANCY_OCID,
+)
 from ads.model import DataScienceModel, ModelVersionSet
 
 logger = logging.getLogger("ads.aqua")
@@ -569,15 +576,13 @@ def get_container_image(
         A dict of allowed configs.
     """
 
+    container_image = UNKNOWN
     config = config_file_name or get_container_config()
     config_file_name = service_config_path()
 
     if container_type not in config:
-        raise AquaValueError(
-            f"{config_file_name} does not have config details for model: {container_type}"
-        )
+        return UNKNOWN
 
-    container_image = None
     mapping = config[container_type]
     versions = [obj["version"] for obj in mapping]
     # assumes numbered versions, update if `latest` is used
@@ -1071,7 +1076,6 @@ def list_hf_models(query: str) -> List[str]:
     try:
         models = HfApi().list_models(
             model_name=query,
-            task="text-generation",
             sort="downloads",
             direction=-1,
             limit=20,
@@ -1079,3 +1083,76 @@ def list_hf_models(query: str) -> List[str]:
         return [model.id for model in models if model.disabled is None]
     except HfHubHTTPError as err:
         raise format_hf_custom_error_message(err) from err
+
+
+def generate_tei_cmd_var(os_path: str) -> List[str]:
+    """This utility functions generates CMD params for Text Embedding Inference container. Only the
+    essential parameters for OCI model deployment are added, defaults are used for the rest.
+    Parameters
+    ----------
+    os_path: str
+        OCI bucket path where the model artifacts are uploaded - oci://bucket@namespace/prefix
+
+    Returns
+    -------
+        cmd_var:
+            List of command line arguments
+    """
+
+    cmd_prefix = "--"
+    cmd_var = [
+        f"{cmd_prefix}{TextEmbeddingInferenceContainerParams.MODEL_ID}",
+        f"{AQUA_MODEL_DEPLOYMENT_FOLDER}{ObjectStorageDetails.from_path(os_path.rstrip('/')).filepath}/",
+        f"{cmd_prefix}{TextEmbeddingInferenceContainerParams.PORT}",
+        TEI_CONTAINER_DEFAULT_HOST,
+    ]
+
+    return cmd_var
+
+
+def parse_cmd_var(cmd_list: List[str]) -> dict:
+    """Helper functions that parses a list into a key-value dictionary. The list contains keys separated by the prefix
+    '--' and the value of the key is the subsequent element.
+    """
+    parsed_cmd = {}
+
+    for i, cmd in enumerate(cmd_list):
+        if cmd.startswith("--"):
+            if i + 1 < len(cmd_list) and not cmd_list[i + 1].startswith("--"):
+                parsed_cmd[cmd] = cmd_list[i + 1]
+                i += 1
+            else:
+                parsed_cmd[cmd] = None
+    return parsed_cmd
+
+
+def validate_cmd_var(cmd_var: List[str], overrides: List[str]) -> List[str]:
+    """This function accepts two lists of parameters and combines them. If the second list shares the common parameter
+    names/keys, then it raises an error.
+    Parameters
+    ----------
+    cmd_var: List[str]
+        Default list of parameters
+    overrides: List[str]
+        List of parameters to override
+    Returns
+    -------
+        List[str] of combined parameters
+    """
+    cmd_var = [str(x) for x in cmd_var]
+    if not overrides:
+        return cmd_var
+    overrides = [str(x) for x in overrides]
+
+    cmd_dict = parse_cmd_var(cmd_var)
+    overrides_dict = parse_cmd_var(overrides)
+
+    # check for conflicts
+    common_keys = set(cmd_dict.keys()) & set(overrides_dict.keys())
+    if common_keys:
+        raise AquaValueError(
+            f"The following CMD input cannot be overridden for model deployment: {', '.join(common_keys)}"
+        )
+
+    combined_cmd_var = cmd_var + overrides
+    return combined_cmd_var

ads/aqua/config/config.py

@@ -29,19 +29,3 @@ def get_evaluation_service_config(
         .get(ContainerSpec.CONTAINER_SPEC, {})
         .get(container, {})
     )
-
-
-# TODO: move this to global config.json in object storage
-def get_finetuning_config_defaults():
-    """Generate and return the fine-tuning default configuration dictionary."""
-    return {
-        "shape": {
-            "VM.GPU.A10.1": {"batch_size": 1, "replica": "1-10"},
-            "VM.GPU.A10.2": {"batch_size": 1, "replica": "1-10"},
-            "BM.GPU.A10.4": {"batch_size": 1, "replica": 1},
-            "BM.GPU4.8": {"batch_size": 4, "replica": 1},
-            "BM.GPU.L40S-NC.4": {"batch_size": 4, "replica": 1},
-            "BM.GPU.A100-v2.8": {"batch_size": 6, "replica": 1},
-            "BM.GPU.H100.8": {"batch_size": 6, "replica": 1},
-        }
-    }

ads/aqua/constants.py

@@ -27,6 +27,7 @@ NB_SESSION_IDENTIFIER = "NB_SESSION_OCID"
 LIFECYCLE_DETAILS_MISSING_JOBRUN = "The associated JobRun resource has been deleted."
 READY_TO_DEPLOY_STATUS = "ACTIVE"
 READY_TO_FINE_TUNE_STATUS = "TRUE"
+PRIVATE_ENDPOINT_TYPE = "MODEL_DEPLOYMENT"
 AQUA_GA_LIST = ["id19sfcrra6z"]
 AQUA_MODEL_TYPE_SERVICE = "service"
 AQUA_MODEL_TYPE_CUSTOM = "custom"
@@ -79,3 +80,4 @@ LLAMA_CPP_INFERENCE_RESTRICTED_PARAMS = {
     "--port",
     "--host",
 }
+TEI_CONTAINER_DEFAULT_HOST = "8080"

ads/aqua/extension/deployment_handler.py

@@ -102,6 +102,7 @@ class AquaDeploymentHandler(AquaAPIhandler):
         ocpus = input_data.get("ocpus")
         memory_in_gbs = input_data.get("memory_in_gbs")
         model_file = input_data.get("model_file")
+        private_endpoint_id = input_data.get("private_endpoint_id")
 
         self.finish(
             AquaDeploymentApp().create(
@@ -124,6 +125,7 @@ class AquaDeploymentHandler(AquaAPIhandler):
                 ocpus=ocpus,
                 memory_in_gbs=memory_in_gbs,
                 model_file=model_file,
+                private_endpoint_id=private_endpoint_id,
             )
         )
 

ads/aqua/extension/finetune_handler.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
@@ -9,8 +8,8 @@ from urllib.parse import urlparse
 from tornado.web import HTTPError
 
 from ads.aqua.common.decorator import handle_exceptions
-from ads.aqua.extension.errors import Errors
 from ads.aqua.extension.base_handler import AquaAPIhandler
+from ads.aqua.extension.errors import Errors
 from ads.aqua.extension.utils import validate_function_parameters
 from ads.aqua.finetuning import AquaFineTuningApp
 from ads.aqua.finetuning.entities import CreateFineTuningDetails

ads/aqua/extension/ui_handler.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
@@ -10,8 +9,9 @@ from tornado.web import HTTPError
 
 from ads.aqua.common.decorator import handle_exceptions
 from ads.aqua.common.enums import Tags
-from ads.aqua.extension.errors import Errors
+from ads.aqua.constants import PRIVATE_ENDPOINT_TYPE
 from ads.aqua.extension.base_handler import AquaAPIhandler
+from ads.aqua.extension.errors import Errors
 from ads.aqua.extension.utils import validate_function_parameters
 from ads.aqua.model.entities import ImportModelDetails
 from ads.aqua.ui import AquaUIApp
@@ -72,6 +72,8 @@ class AquaUIHandler(AquaAPIhandler):
             return self.list_vcn()
         elif paths.startswith("aqua/subnets"):
             return self.list_subnets()
+        elif paths.startswith("aqua/privateendpoints"):
+            return self.list_private_endpoints()
         elif paths.startswith("aqua/shapes/limit"):
             return self.get_shape_availability()
         elif paths.startswith("aqua/bucket/versioning"):
@@ -175,15 +177,31 @@ class AquaUIHandler(AquaAPIhandler):
             )
         )
 
+    def list_private_endpoints(self, **kwargs):
+        """Lists the private endpoints in the specified compartment."""
+        compartment_id = self.get_argument("compartment_id", default=COMPARTMENT_OCID)
+        resource_type = self.get_argument(
+            "resource_type", default=PRIVATE_ENDPOINT_TYPE
+        )
+        return self.finish(
+            AquaUIApp().list_private_endpoints(
+                compartment_id=compartment_id, resource_type=resource_type, **kwargs
+            )
+        )
+
     def get_shape_availability(self, **kwargs):
         """For a given compartmentId, resource limit name, and scope, returns the number of available resources associated
         with the given limit."""
         compartment_id = self.get_argument("compartment_id", default=COMPARTMENT_OCID)
         instance_shape = self.get_argument("instance_shape")
+        limit_name = self.get_argument("limit_name")
 
         return self.finish(
             AquaUIApp().get_shape_availability(
-                compartment_id=compartment_id, instance_shape=instance_shape, **kwargs
+                compartment_id=compartment_id,
+                instance_shape=instance_shape,
+                limit_name=limit_name,
+                **kwargs,
             )
         )
 
@@ -244,6 +262,7 @@ __handlers__ = [
     ("job/shapes/?([^/]*)", AquaUIHandler),
     ("vcn/?([^/]*)", AquaUIHandler),
     ("subnets/?([^/]*)", AquaUIHandler),
+    ("privateendpoints/?([^/]*)", AquaUIHandler),
     ("shapes/limit/?([^/]*)", AquaUIHandler),
     ("bucket/versioning/?([^/]*)", AquaUIHandler),
     ("containers/?([^/]*)", AquaUIHandler),

ads/aqua/finetuning/entities.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 from dataclasses import dataclass, field
@@ -14,9 +13,9 @@ class AquaFineTuningParams(DataClassSerializable):
     epochs: int
     learning_rate: Optional[float] = None
     sample_packing: Optional[bool] = "auto"
-    batch_size: Optional[
-        int
-    ] = None  # make it batch_size for user, but internally this is micro_batch_size
+    batch_size: Optional[int] = (
+        None  # make it batch_size for user, but internally this is micro_batch_size
+    )
     sequence_len: Optional[int] = None
     pad_to_sequence_len: Optional[bool] = None
     lora_r: Optional[int] = None
@@ -24,6 +23,8 @@ class AquaFineTuningParams(DataClassSerializable):
     lora_dropout: Optional[float] = None
     lora_target_linear: Optional[bool] = None
     lora_target_modules: Optional[List] = None
+    early_stopping_patience: Optional[int] = None
+    early_stopping_threshold: Optional[float] = None
 
 
 @dataclass(repr=False)

ads/aqua/finetuning/finetuning.py

@@ -1,11 +1,10 @@
 #!/usr/bin/env python
-# -*- coding: utf-8 -*-
 # Copyright (c) 2024 Oracle and/or its affiliates.
 # Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
 
 import json
 import os
-from dataclasses import asdict, fields, MISSING
+from dataclasses import MISSING, asdict, fields
 from typing import Dict
 
 from oci.data_science.models import (
@@ -14,7 +13,7 @@ from oci.data_science.models import (
     UpdateModelProvenanceDetails,
 )
 
-from ads.aqua import ODSC_MODEL_COMPARTMENT_OCID, logger
+from ads.aqua import logger
 from ads.aqua.app import AquaApp
 from ads.aqua.common.enums import Resource, Tags
 from ads.aqua.common.errors import AquaFileExistsError, AquaValueError
@@ -31,7 +30,6 @@ from ads.aqua.constants import (
     UNKNOWN,
     UNKNOWN_DICT,
 )
-from ads.aqua.config.config import get_finetuning_config_defaults
 from ads.aqua.data import AquaResourceIdentifier
 from ads.aqua.finetuning.constants import *
 from ads.aqua.finetuning.entities import *
@@ -132,7 +130,7 @@ class AquaFineTuningApp(AquaApp):
             or create_fine_tuning_details.validation_set_size >= 1
         ):
             raise AquaValueError(
-                f"Fine tuning validation set size should be a float number in between [0, 1)."
+                "Fine tuning validation set size should be a float number in between [0, 1)."
             )
 
         if create_fine_tuning_details.replica < DEFAULT_FT_REPLICA:
@@ -394,7 +392,7 @@ class AquaFineTuningApp(AquaApp):
         )
         # track shapes that were used for fine-tune creation
         self.telemetry.record_event_async(
-            category=f"aqua/service/finetune/create/shape/",
+            category="aqua/service/finetune/create/shape/",
             action=f"{create_fine_tuning_details.shape_name}x{create_fine_tuning_details.replica}",
             **telemetry_kwargs,
         )
@@ -533,6 +531,12 @@ class AquaFineTuningApp(AquaApp):
                     oci_launch_cmd += f"--num_{key} {value} "
                 elif key == "lora_target_modules":
                     oci_launch_cmd += f"--{key} {','.join(str(k) for k in value)} "
+                elif key == "early_stopping_patience":
+                    if value != 0:
+                        oci_launch_cmd += f"--{key} {value} "
+                elif key == "early_stopping_threshold":
+                    if "early_stopping_patience" in oci_launch_cmd:
+                        oci_launch_cmd += f"--{key} {value} "
                 else:
                     oci_launch_cmd += f"--{key} {value} "
 
@@ -558,8 +562,9 @@ class AquaFineTuningApp(AquaApp):
 
         config = self.get_config(model_id, AQUA_MODEL_FINETUNING_CONFIG)
         if not config:
-            logger.info(f"Fetching default fine-tuning config for model: {model_id}")
-            config = get_finetuning_config_defaults()
+            logger.debug(
+                f"Fine-tuning config for custom model: {model_id} is not available."
+            )
         return config
 
     @telemetry(

ads/aqua/model/constants.py

@@ -17,6 +17,7 @@ class ModelCustomMetadataFields(str, metaclass=ExtendedEnumMeta):
     DEPLOYMENT_CONTAINER = "deployment-container"
     EVALUATION_CONTAINER = "evaluation-container"
     FINETUNE_CONTAINER = "finetune-container"
+    DEPLOYMENT_CONTAINER_URI = "deployment-container-uri"
 
 
 class ModelTask(str, metaclass=ExtendedEnumMeta):

ads/aqua/model/entities.py

@@ -98,6 +98,7 @@ class AquaModel(AquaModelSummary, DataClassSerializable):
 
     model_card: str = None
     inference_container: str = None
+    inference_container_uri: str = None
     finetuning_container: str = None
     evaluation_container: str = None
     artifact_location: str = None
@@ -287,6 +288,7 @@ class ImportModelDetails(CLIBuilderMixin):
     compartment_id: Optional[str] = None
     project_id: Optional[str] = None
     model_file: Optional[str] = None
+    inference_container_uri: Optional[str] = None
 
     def __post_init__(self):
         self._command = "model register"