Merge branch 'main' into feature/aqua-v1.0.5

Author: kumar-shivam-ranjan

.github/workflows/run-forecast-unit-tests.yml

@@ -56,7 +56,6 @@ jobs:
           $CONDA/bin/conda init
           source /home/runner/.bashrc
           pip install -r test-requirements-operators.txt
-          pip install "oracle-automlx[classic]>=24.2.0"
-          pip install "oracle-automlx[forecasting]>=24.2.0"
+          pip install "oracle-automlx[forecasting]>=24.4.0"
           pip install pandas>=2.2.0
           python -m pytest -v -p no:warnings --durations=5 tests/operators/forecast

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 @@
     InferenceContainerParamType,
     InferenceContainerType,
     RqsAdditionalDetails,
+    TextEmbeddingInferenceContainerParams,
 )
 from ads.aqua.common.errors import (
     AquaFileNotFoundError,
@@ -51,6 +52,7 @@
     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.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
@@ -1078,3 +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/constants.py

@@ -80,3 +80,4 @@
     "--port",
     "--host",
 }
+TEI_CONTAINER_DEFAULT_HOST = "8080"

ads/aqua/evaluation/entities.py

@@ -9,19 +9,18 @@
 This module contains dataclasses for aqua evaluation.
 """
 
-from dataclasses import dataclass, field
-from typing import List, Optional, Union
+from pydantic import Field
+from typing import Any, Dict, List, Optional, Union
 
 from ads.aqua.data import AquaResourceIdentifier
-from ads.common.serializer import DataClassSerializable
+from ads.aqua.config.utils.serializer import Serializable
 
 
-@dataclass(repr=False)
-class CreateAquaEvaluationDetails(DataClassSerializable):
-    """Dataclass to create aqua model evaluation.
+class CreateAquaEvaluationDetails(Serializable):
+    """Class for creating aqua model evaluation.
 
-    Fields
-    ------
+    Properties
+    ----------
     evaluation_source_id: str
         The evaluation source id. Must be either model or model deployment ocid.
     evaluation_name: str
@@ -83,69 +82,64 @@ class CreateAquaEvaluationDetails(DataClassSerializable):
     ocpus: Optional[float] = None
     log_group_id: Optional[str] = None
     log_id: Optional[str] = None
-    metrics: Optional[List] = None
+    metrics: Optional[List[str]] = None
     force_overwrite: Optional[bool] = False
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class AquaEvalReport(DataClassSerializable):
+class AquaEvalReport(Serializable):
     evaluation_id: str = ""
     content: str = ""
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class ModelParams(DataClassSerializable):
-    max_tokens: str = ""
-    top_p: str = ""
-    top_k: str = ""
-    temperature: str = ""
-    presence_penalty: Optional[float] = 0.0
-    frequency_penalty: Optional[float] = 0.0
-    stop: Optional[Union[str, List[str]]] = field(default_factory=list)
-    model: Optional[str] = "odsc-llm"
-
-
-@dataclass(repr=False)
-class AquaEvalParams(ModelParams, DataClassSerializable):
+class AquaEvalParams(Serializable):
     shape: str = ""
     dataset_path: str = ""
     report_path: str = ""
 
+    class Config:
+        extra = "allow"
 
-@dataclass(repr=False)
-class AquaEvalMetric(DataClassSerializable):
+class AquaEvalMetric(Serializable):
     key: str
     name: str
     description: str = ""
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class AquaEvalMetricSummary(DataClassSerializable):
+class AquaEvalMetricSummary(Serializable):
     metric: str = ""
     score: str = ""
     grade: str = ""
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class AquaEvalMetrics(DataClassSerializable):
+class AquaEvalMetrics(Serializable):
     id: str
     report: str
-    metric_results: List[AquaEvalMetric] = field(default_factory=list)
-    metric_summary_result: List[AquaEvalMetricSummary] = field(default_factory=list)
+    metric_results: List[AquaEvalMetric] = Field(default_factory=list)
+    metric_summary_result: List[AquaEvalMetricSummary] = Field(default_factory=list)
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class AquaEvaluationCommands(DataClassSerializable):
+class AquaEvaluationCommands(Serializable):
     evaluation_id: str
     evaluation_target_id: str
-    input_data: dict
-    metrics: list
+    input_data: Dict[str, Any]
+    metrics: List[str]
     output_dir: str
-    params: dict
+    params: Dict[str, Any]
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class AquaEvaluationSummary(DataClassSerializable):
+class AquaEvaluationSummary(Serializable):
     """Represents a summary of Aqua evalution."""
 
     id: str
@@ -154,17 +148,18 @@ class AquaEvaluationSummary(DataClassSerializable):
     lifecycle_state: str
     lifecycle_details: str
     time_created: str
-    tags: dict
-    experiment: AquaResourceIdentifier = field(default_factory=AquaResourceIdentifier)
-    source: AquaResourceIdentifier = field(default_factory=AquaResourceIdentifier)
-    job: AquaResourceIdentifier = field(default_factory=AquaResourceIdentifier)
-    parameters: AquaEvalParams = field(default_factory=AquaEvalParams)
+    tags: Dict[str, Any]
+    experiment: AquaResourceIdentifier = Field(default_factory=AquaResourceIdentifier)
+    source: AquaResourceIdentifier = Field(default_factory=AquaResourceIdentifier)
+    job: AquaResourceIdentifier = Field(default_factory=AquaResourceIdentifier)
+    parameters: AquaEvalParams = Field(default_factory=AquaEvalParams)
 
+    class Config:
+        extra = "ignore"
 
-@dataclass(repr=False)
-class AquaEvaluationDetail(AquaEvaluationSummary, DataClassSerializable):
+class AquaEvaluationDetail(AquaEvaluationSummary):
     """Represents a details of Aqua evalution."""
 
-    log_group: AquaResourceIdentifier = field(default_factory=AquaResourceIdentifier)
-    log: AquaResourceIdentifier = field(default_factory=AquaResourceIdentifier)
-    introspection: dict = field(default_factory=dict)
+    log_group: AquaResourceIdentifier = Field(default_factory=AquaResourceIdentifier)
+    log: AquaResourceIdentifier = Field(default_factory=AquaResourceIdentifier)
+    introspection: dict = Field(default_factory=dict)