eval-studio-client 0.8.0a2__py3-none-any.whl → 1.0.0a1__py3-none-any.whl

eval_studio_client/insights.py

@@ -22,7 +22,7 @@ class Insight:
     _evaluator_id: Optional[str] = None
 
     @staticmethod
-    def _from_api_insight(api_insight: models.V1alphaInsight) -> "Insight":
+    def _from_api_insight(api_insight: models.V1Insight) -> "Insight":
         """Converts an API Insight to a client Insight."""
         return Insight(
             description=api_insight.description or "",

eval_studio_client/leaderboards.py

@@ -33,10 +33,11 @@ class Leaderboard:
     update_time: Optional[datetime.datetime] = None
     problems: List[p6s.Problem] = dataclasses.field(default_factory=list)
     insights: List[i6s.Insight] = dataclasses.field(default_factory=list)
+    existing_collection: Optional[str] = None
     _report: Optional[str] = None
     _leaderboard: Optional[str] = None
     _model_name: Optional[str] = None
-    _status: Optional[models.V1alphaLeaderboardStatus] = None
+    _status: Optional[models.V1LeaderboardStatus] = None
     _client: Optional[api.ApiClient] = None
 
     def __post_init__(self):
@@ -60,16 +61,14 @@ class Leaderboard:
     def finished(self) -> bool:
         """Indicates whether the leaderboard has finished."""
         return self._status in [
-            models.V1alphaLeaderboardStatus.LEADERBOARD_STATUS_COMPLETED,
-            models.V1alphaLeaderboardStatus.LEADERBOARD_STATUS_FAILED,
+            models.V1LeaderboardStatus.LEADERBOARD_STATUS_COMPLETED,
+            models.V1LeaderboardStatus.LEADERBOARD_STATUS_FAILED,
         ]
 
     @property
     def successful(self) -> bool:
         """Indicates whether the leaderboard has finished successfully."""
-        return (
-            self._status == models.V1alphaLeaderboardStatus.LEADERBOARD_STATUS_COMPLETED
-        )
+        return self._status == models.V1LeaderboardStatus.LEADERBOARD_STATUS_COMPLETED
 
     @property
     def test_suite(self) -> List[tests.Test]:
@@ -154,18 +153,19 @@ class Leaderboard:
 
         raise TimeoutError("Waiting timeout has been reached.")
 
-    def to_api_proto(self) -> models.V1alphaLeaderboard:
+    def to_api_proto(self) -> models.V1Leaderboard:
         """Converts the client Leaderboard to an API Leaderboard."""
-        return models.V1alphaLeaderboard(
+        return models.V1Leaderboard(
             display_name=self.name,
             description=self.description,
             llm_models=self.base_models or None,
             evaluator=self._evaluator_name,
             tests=self._test_names,
             model=self._model_name,
+            h2ogpte_collection=self.existing_collection or None,
         )
 
-    def _update_result(self, api_leaderboard: models.V1alphaLeaderboard):
+    def _update_result(self, api_leaderboard: models.V1Leaderboard):
         """Refresh the leaderboard with the latest API data."""
         self.key = api_leaderboard.name or ""
         self.update_time = api_leaderboard.update_time
@@ -175,7 +175,7 @@ class Leaderboard:
 
     @staticmethod
     def _from_api_leaderboard(
-        api_leaderboard: models.V1alphaLeaderboard, client: Optional[api.ApiClient]
+        api_leaderboard: models.V1Leaderboard, client: Optional[api.ApiClient]
     ) -> "Leaderboard":
         """Converts an API Leaderboard to a client Leaderboard."""
         api_problems = api_leaderboard.leaderboard_problems or []
@@ -191,6 +191,7 @@ class Leaderboard:
             update_time=api_leaderboard.update_time,
             problems=problems,
             insights=insights,
+            existing_collection=api_leaderboard.h2ogpte_collection or None,
             _evaluator_name=api_leaderboard.evaluator or "",
             _test_names=api_leaderboard.tests or [],
             _report=api_leaderboard.leaderboard_report or "",
@@ -200,8 +201,8 @@ class Leaderboard:
         )
 
     @staticmethod
-    def _is_finished_leaderboard(leaderboard: models.V1alphaLeaderboard) -> bool:
+    def _is_finished_leaderboard(leaderboard: models.V1Leaderboard) -> bool:
         return leaderboard.status in [
-            models.V1alphaLeaderboardStatus.LEADERBOARD_STATUS_COMPLETED,
-            models.V1alphaLeaderboardStatus.LEADERBOARD_STATUS_FAILED,
+            models.V1LeaderboardStatus.LEADERBOARD_STATUS_COMPLETED,
+            models.V1LeaderboardStatus.LEADERBOARD_STATUS_FAILED,
         ]

eval_studio_client/models.py

@@ -15,6 +15,34 @@ from eval_studio_client.api import models
 # Key for Azure environment ID parameter within Model parameters.
 _AZURE_ENV_ID_PARAM = "environment_id"
 
+# Resource name of the default RAG model.
+DEFAULT_RAG_MODEL_KEY = "models/defaultRAGModel"
+
+# Resource name of the default LLM model.
+DEFAULT_LLM_MODEL_KEY = "models/defaultLLMModel"
+
+
+@dataclasses.dataclass
+class CollectionInfo:
+    """Represents the information about a collection in the H2OGPTE
+    or a Knowledge Base in Amazon Bedrock.
+    """
+
+    key: str
+    name: str
+    description: str
+
+    def __str__(self):
+        return f"{self.name} ({self.key})"
+
+    @staticmethod
+    def _from_api_collection_info(api_col: models.V1CollectionInfo) -> "CollectionInfo":
+        return CollectionInfo(
+            key=api_col.id or "",
+            name=api_col.display_name or "",
+            description=api_col.description or "",
+        )
+
 
 @dataclasses.dataclass
 class Model:
@@ -55,7 +83,7 @@ class Model:
         while True:
             res = self._leaderboard_api.leaderboard_service_list_leaderboards(
                 filter=f'model="{self.key}"',
-                view=models.V1alphaLeaderboardView.LEADERBOARD_VIEW_BASIC_WITH_TABLE,
+                view=models.V1LeaderboardView.LEADERBOARD_VIEW_BASIC_WITH_TABLE,
                 page_token=page_token,
             )
             if not res or not res.leaderboards:
@@ -74,6 +102,28 @@ class Model:
 
         return result
 
+    @property
+    def base_models(self) -> List[str]:
+        """List of base LLM models available to use e.g. for the evaluation."""
+        res = self._model_api.model_service_list_base_models(self.key)
+        if res and res.base_models:
+            return [str(m) for m in res.base_models]
+
+        raise RuntimeError("Failed to list base models")
+
+    @property
+    def collections(self) -> List[CollectionInfo]:
+        """List of collections available for evaluation.
+
+        NOTE: This is currently supported only for H2OGPTe and Amazon Bedrock RAG
+        model hosts.
+        """
+        res = self._model_api.model_service_list_model_collections(self.key)
+        if res and res.collections:
+            return list(res.collections)
+
+        raise RuntimeError("Failed to list model host collections")
+
     def create_leaderboard(
         self,
         name: str,
@@ -82,20 +132,27 @@ class Model:
         description: Optional[str] = None,
         base_models: Optional[List[str]] = None,
         use_cache: bool = True,
+        existing_collection: Optional[str] = None,
     ) -> Optional[l10s.Leaderboard]:
         """Runs a new evaluation for the model and creates a new leaderboard.
 
         Args:
+            name: The name of the leaderboard.
             evaluator: The evaluator to use for the evaluation.
             test_suite: The list of tests used to evaluate the model.
+            description (optional): The description of the leaderboard.
             base_models (optional): The base LLM models to use for the evaluation.
             use_cache (optional): Whether to use the cached answers if available.
+            existing_collection (str): ID or the resource name of the existing
+                collection, which will be used as a corpus for evaluation.
+                NOTE: This option works only for the H2OGPTe and Amazon Bedrock model hosts ATM.
         """
         lb = l10s.Leaderboard(
             key="",
             name=name,
             description=description or "",
             base_models=base_models or [],
+            existing_collection=existing_collection,
             _model_name=self.key,
             _evaluator_name=evaluator.key,
             _test_names=[t.key for t in test_suite],
@@ -122,6 +179,7 @@ class Model:
         test_suites: Union[tests.Test, List[tests.Test]],
         description: Optional[str] = None,
         base_models: Optional[List[str]] = None,
+        existing_collection: Optional[str] = None,
     ) -> Optional[d8s.Dashboard]:
         """Runs a new evaluation for the model and creates a new dashboard.
 
@@ -130,6 +188,9 @@ class Model:
             test_suites: The test(s) used to evaluate the model.
             description (optional): The description of the dashboard.
             base_models (optional): The base LLM models to use for the evaluation.
+            existing_collection (str): ID or the resource name of the existing
+                collection, which will be used as a corpus for evaluation.
+                NOTE: This option works only for the H2OGPTe and Amazon Bedrock model hosts ATM.
         """
         _evaluators = (
             [evaluators] if isinstance(evaluators, e8s.Evaluator) else evaluators
@@ -138,25 +199,26 @@ class Model:
             [test_suites] if isinstance(test_suites, tests.Test) else test_suites
         )
 
-        create_lb_reqs: List[models.V1alphaCreateLeaderboardRequest] = []
+        create_lb_reqs: List[models.V1CreateLeaderboardRequest] = []
         for evaluator in _evaluators:
             lb = l10s.Leaderboard(
                 key="",
                 name=f"{name} - {evaluator.name}",
                 description=description or "",
                 base_models=base_models or [],
+                existing_collection=existing_collection,
                 _model_name=self.key,
                 _evaluator_name=evaluator.key,
                 _test_names=[t.key for t in _test_suites],
                 _client=self._client,
             )
-            create_lb_req = models.V1alphaCreateLeaderboardRequest(
+            create_lb_req = models.V1CreateLeaderboardRequest(
                 leaderboard=lb.to_api_proto()
             )
             create_lb_reqs.append(create_lb_req)
 
         res = self._leaderboard_api.leaderboard_service_batch_create_leaderboards(
-            models.V1alphaBatchCreateLeaderboardsRequest(
+            models.V1BatchCreateLeaderboardsRequest(
                 requests=create_lb_reqs,
                 dashboard_display_name=name,
                 dashboard_description=description,
@@ -184,7 +246,7 @@ class Model:
             test_lab: The test lab in JSON format to use for the evaluation.
             description (optional): The description of the leaderboard.
         """
-        req = models.V1alphaImportLeaderboardRequest(
+        req = models.V1ImportLeaderboardRequest(
             testLabJson=test_lab,
             evaluator=evaluator.key,
             model=self.key,
@@ -212,7 +274,7 @@ class Model:
         raise RuntimeError("Failed to list base models")
 
     def _get_leaderboard_from_operation(
-        self, operation: models.V1alphaOperation
+        self, operation: models.V1Operation
     ) -> Optional[l10s.Leaderboard]:
         """Retrieves the leaderboard from the operation, which created it.
 
@@ -230,7 +292,7 @@ class Model:
         return None
 
     def _get_dashboard_from_operation(
-        self, operation: models.V1alphaOperation
+        self, operation: models.V1Operation
     ) -> Optional[d8s.Dashboard]:
         """Retrieves the dashboard from the operation, which created it.
 
@@ -251,9 +313,7 @@ class Model:
         return None
 
     @staticmethod
-    def _from_api_model(
-        api_model: models.V1alphaModel, client: api.ApiClient
-    ) -> "Model":
+    def _from_api_model(api_model: models.V1Model, client: api.ApiClient) -> "Model":
         """Converts the API model to the client model."""
         return Model(
             key=api_model.name or "",
@@ -268,10 +328,10 @@ class Model:
         )
 
     @staticmethod
-    def _is_rag_model(api_model: models.V1alphaModel) -> bool:
+    def _is_rag_model(api_model: models.V1Model) -> bool:
         return api_model.type in [
-            models.V1alphaModelType.MODEL_TYPE_H2_OGPTE_RAG,
-            models.V1alphaModelType.MODEL_TYPE_OPENAI_RAG,
+            models.V1ModelType.MODEL_TYPE_H2_OGPTE_RAG,
+            models.V1ModelType.MODEL_TYPE_OPENAI_RAG,
         ]
 
 
@@ -285,6 +345,12 @@ class _Models:
 
         Args:
             key: The model resource name to retrieve.
+
+        Returns:
+            Model: The model object.
+
+        Raises:
+            KeyError: If the model is not found.
         """
         res = self._api.model_service_get_model(key)
         if res and res.model:
@@ -292,6 +358,28 @@ class _Models:
 
         raise KeyError("Model not found.")
 
+    def get_default_rag(self) -> Model:
+        """Gets the default RAG model from Eval Studio.
+
+        Returns:
+            Model: The default RAG model object.
+
+        Raises:
+            KeyError: If no default RAG model is set.
+        """
+        return self.get(DEFAULT_RAG_MODEL_KEY)
+
+    def get_default_llm(self) -> Model:
+        """Gets the default LLM model from Eval Studio.
+
+        Returns:
+            Model: The default LLM model object.
+
+        Raises:
+            KeyError: If no default LLM model is set.
+        """
+        return self.get(DEFAULT_LLM_MODEL_KEY)
+
     def create_h2ogpte_model(
         self, name: str, is_rag: bool, description: str, url: str, api_key: str
     ) -> Model:
@@ -308,11 +396,11 @@ class _Models:
             api_key: API key for the model host system.
         """
         model_type = (
-            models.V1alphaModelType.MODEL_TYPE_H2_OGPTE_RAG
+            models.V1ModelType.MODEL_TYPE_H2_OGPTE_RAG
             if is_rag
-            else models.V1alphaModelType.MODEL_TYPE_H2_OGPTE_LLM
+            else models.V1ModelType.MODEL_TYPE_H2_OGPTE_LLM
         )
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
             url=url,
@@ -336,12 +424,12 @@ class _Models:
             url: URL of the model host system.
             api_key: API key for the model host system.
         """
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
             url=url,
             api_key=api_key,
-            type=models.V1alphaModelType.MODEL_TYPE_H2_OGPT_LLM,
+            type=models.V1ModelType.MODEL_TYPE_H2_OGPT_LLM,
         )
         res = self._api.model_service_create_model(req)
         if res and res.model:
@@ -360,12 +448,12 @@ class _Models:
             url: URL of the model host system.
             api_key: API key for the model host system.
         """
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
             url=url,
             api_key=api_key,
-            type=models.V1alphaModelType.MODEL_TYPE_H2_OLLMOPS,
+            type=models.V1ModelType.MODEL_TYPE_H2_OLLMOPS,
         )
         res = self._api.model_service_create_model(req)
         if res and res.model:
@@ -399,11 +487,11 @@ class _Models:
             )
 
         model_type = (
-            models.V1alphaModelType.MODEL_TYPE_OPENAI_RAG
+            models.V1ModelType.MODEL_TYPE_OPENAI_RAG
             if is_rag
-            else models.V1alphaModelType.MODEL_TYPE_OPENAI_CHAT
+            else models.V1ModelType.MODEL_TYPE_OPENAI_CHAT
         )
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
             api_key=api_key,
@@ -429,12 +517,12 @@ class _Models:
             environmentID: Azure environment ID.
         """
         params = {_AZURE_ENV_ID_PARAM: environmentID}
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
             url=url,
             api_key=api_key,
-            type=models.V1alphaModelType.MODEL_TYPE_AZURE_OPENAI_CHAT,
+            type=models.V1ModelType.MODEL_TYPE_AZURE_OPENAI_CHAT,
             parameters=json.dumps(params),
         )
         res = self._api.model_service_create_model(req)
@@ -454,12 +542,12 @@ class _Models:
             url: URL of the model host system.
             api_key: API key for the model host system.
         """
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
             url=url,
             api_key=api_key,
-            type=models.V1alphaModelType.MODEL_TYPE_OLLAMA,
+            type=models.V1ModelType.MODEL_TYPE_OLLAMA,
         )
         res = self._api.model_service_create_model(req)
         if res and res.model:
@@ -491,10 +579,10 @@ class _Models:
             "aws_secret_access_key": aws_secret_access_key,
             "aws_session_token": aws_session_token,
         }
-        req = models.V1alphaModel(
+        req = models.V1Model(
             display_name=name,
             description=description,
-            type=models.V1alphaModelType.MODEL_TYPE_AMAZON_BEDROCK,
+            type=models.V1ModelType.MODEL_TYPE_AMAZON_BEDROCK,
             api_key=json.dumps(credentials),
             parameters=json.dumps({"region": aws_region}),
         )

eval_studio_client/perturbators.py

@@ -17,12 +17,12 @@ class PerturbatorIntensity(enum.Enum):
     medium = "medium"
     high = "high"
 
-    def to_api_proto(self) -> models.V1alphaPerturbatorIntensity:
+    def to_api_proto(self) -> models.V1PerturbatorIntensity:
         """Converts the client PerturbatorIntensity to an API PerturbatorIntensity."""
         proto_values = {
-            PerturbatorIntensity.low: models.V1alphaPerturbatorIntensity.PERTURBATOR_INTENSITY_LOW,
-            PerturbatorIntensity.medium: models.V1alphaPerturbatorIntensity.PERTURBATOR_INTENSITY_MEDIUM,
-            PerturbatorIntensity.high: models.V1alphaPerturbatorIntensity.PERTURBATOR_INTENSITY_HIGH,
+            PerturbatorIntensity.low: models.V1PerturbatorIntensity.PERTURBATOR_INTENSITY_LOW,
+            PerturbatorIntensity.medium: models.V1PerturbatorIntensity.PERTURBATOR_INTENSITY_MEDIUM,
+            PerturbatorIntensity.high: models.V1PerturbatorIntensity.PERTURBATOR_INTENSITY_HIGH,
         }
 
         return proto_values[self]
@@ -52,9 +52,7 @@ class Perturbator:
         self.intensity = self.intensity or PerturbatorIntensity.medium
 
     @staticmethod
-    def _from_api_perturbator(
-        api_perturbator: models.V1alphaPerturbator,
-    ) -> "Perturbator":
+    def _from_api_perturbator(api_perturbator: models.V1Perturbator) -> "Perturbator":
         """Converts an API Perturbator to a client Perturbator."""
         return Perturbator(
             key=api_perturbator.name or "",

eval_studio_client/problems.py

@@ -32,7 +32,7 @@ class Problem:
     _evaluator_id: Optional[str] = None
 
     @staticmethod
-    def _from_api_problem(api_problem: models.V1alphaProblemAndAction) -> "Problem":
+    def _from_api_problem(api_problem: models.V1ProblemAndAction) -> "Problem":
         """Converts an API Problem to a client Problem."""
         try:
             severity = ProblemSeverity(api_problem.severity)

eval_studio_client/test_labs.py

@@ -98,7 +98,7 @@ class TestLab:
         Args:
             evaluator: The evaluator to use for the evaluation.
         """
-        req = apiModels.V1alphaImportLeaderboardRequest(
+        req = apiModels.V1ImportLeaderboardRequest(
             testLabJson=self.json(),
             evaluator=evaluator.key,
             model=None,
@@ -132,7 +132,7 @@ class TestLab:
         return json.dumps(lab, indent=4, sort_keys=True)
 
     def _get_leaderboard_from_operation(
-        self, operation: apiModels.V1alphaOperation
+        self, operation: apiModels.V1Operation
     ) -> Optional[l10s.Leaderboard]:
         """Retrieves the leaderboard from the operation, which created it.