eval-studio-client 0.8.2__py3-none-any.whl → 1.0.0__py3-none-any.whl

eval_studio_client/dashboards.py

@@ -9,6 +9,7 @@ from eval_studio_client import api
 from eval_studio_client import insights as i6s
 from eval_studio_client import leaderboards as l10s
 from eval_studio_client import problems as p6s
+from eval_studio_client import utils
 from eval_studio_client.api import models
 
 
@@ -41,6 +42,7 @@ class Dashboard:
             self._dashboard_api = api.DashboardServiceApi(self._client)
             self._leaderboard_api = api.LeaderboardServiceApi(self._client)
             self._info_api = api.InfoServiceApi(self._client)
+            self._operation_api = api.OperationServiceApi(self._client)
 
     @property
     def leaderboards(self) -> Optional[List[l10s.Leaderboard]]:
@@ -118,36 +120,56 @@ class Dashboard:
         else:
             raise ValueError("Cannot establish connection to Eval Studio host.")
 
-    def wait_to_finish(self, timeout: Optional[float] = None):
+    def wait_to_finish(self, timeout: Optional[float] = None, verbose: bool = False):
         """Waits for the dashboard to finish.
 
         Args:
             timeout: The maximum time to wait in seconds.
+            verbose (bool): If True, prints the status of the evaluation while waiting.
         """
         timeout = timeout or float("inf")
+        progress_bar = utils.ProgressBar()
         if self.finished:
             return
 
+        if not self._create_operation:
+            # This means that the evaluation has no assigned operation, thus cannot poll.
+            raise RuntimeError("Failed to retrieve running evaluation info.")
+
         if self._client:
             ctr = 0
             while ctr < timeout:
-                lbs = self.leaderboards
-                if lbs:
-                    if all(lb.finished for lb in lbs):
-                        return
-
-                    ctr += 1
-                    time.sleep(1)
+                op = self._operation_api.operation_service_get_operation(
+                    self._create_operation
+                )
+                if not op or not op.operation:
+                    raise RuntimeError(
+                        "Failed to retrieve running evaluation progress."
+                    )
+
+                if verbose:
+                    if not op.operation.metadata:
+                        raise RuntimeError(
+                            "Failed to retrieve running evaluation progress details."
+                        )
+
+                    op_meta = op.operation.metadata.to_dict()
+                    progress = op_meta.get("progress", 0)
+                    progress_msg = op_meta.get("progressMessage", "Running")
+                    progress_bar.update(progress, progress_msg)
+
+                if op.operation.done:
+                    return
+
+                ctr += 1
+                time.sleep(1)
         else:
             raise ValueError("Cannot establish connection to Eval Studio host.")
 
         raise TimeoutError("Waiting timeout has been reached.")
 
-    def show(self):
-        """Opens the evaluation in the default web browser.
-
-        NOTE: This functionality is primarily for interactive use in Jupyter notebooks.
-        """
+    def show(self) -> str:
+        """Prints the endpoint URL of the evaluation dashboard."""
         if self._client:
             info_res = self._info_api.info_service_get_info()
             if not info_res or not info_res.info:
@@ -155,11 +177,8 @@ class Dashboard:
 
             host = info_res.info.base_url
             url = urllib.parse.urljoin(host, self.key)
-
-            # NOTE: Local import is used to avoid problems for users outside Jupyter environment.
-            import webbrowser
-
-            webbrowser.open(url)
+            print(f"Open following url to access evaluation dashboard: \n\n{url}")
+            return url
         else:
             raise ValueError("Cannot establish connection to Eval Studio host.")
 

eval_studio_client/leaderboards.py

@@ -33,11 +33,13 @@ 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.V1LeaderboardStatus] = None
     _client: Optional[api.ApiClient] = None
+    _operation: Optional[str] = None
 
     def __post_init__(self):
         self._evaluator_api = api.EvaluatorServiceApi(self._client)
@@ -161,6 +163,7 @@ class Leaderboard:
             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.V1Leaderboard):
@@ -189,12 +192,14 @@ 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 "",
             _leaderboard=api_leaderboard.leaderboard_table,
             _status=api_leaderboard.status,
             _client=client,
+            _operation=api_leaderboard.create_operation or None,
         )
 
     @staticmethod

eval_studio_client/models.py

@@ -22,6 +22,28 @@ DEFAULT_RAG_MODEL_KEY = "models/defaultRAGModel"
 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:
     """Represents Eval Studio connection to an external RAG/LLM system.
@@ -80,6 +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,
@@ -88,20 +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],
@@ -128,6 +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.
 
@@ -136,6 +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
@@ -151,6 +206,7 @@ class Model:
                 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],

eval_studio_client/tests.py

@@ -11,6 +11,7 @@ from typing import Union
 from eval_studio_client import api
 from eval_studio_client import documents as d7s
 from eval_studio_client import perturbators as p10s
+from eval_studio_client import utils
 from eval_studio_client.api import models
 
 
@@ -85,15 +86,9 @@ class TestCaseGenerator(enum.Enum):
 
 
 @dataclasses.dataclass
-class TestCaseGenerationHandle:
+class _TestCaseGenerationHandle:
 
     name: Any | None
-    create_time: Optional[datetime.datetime] = None
-    creator: Optional[str] = None
-    update_time: Optional[datetime.datetime] = None
-    updater: Optional[str] = None
-    delete_time: Optional[datetime.datetime] = None
-    deleter: Optional[str] = None
     progress: Optional[float] = None
     progress_message: Optional[str] = None
     error: Optional[models.RpcStatus] = None
@@ -102,11 +97,11 @@ class TestCaseGenerationHandle:
     @staticmethod
     def _from_operation(
         res: models.V1GenerateTestCasesResponse | models.V1GetOperationResponse,
-    ) -> "TestCaseGenerationHandle":
+    ) -> "_TestCaseGenerationHandle":
         """Converts an API operation to prompt generation handle."""
         op: models.V1Operation | None = res.operation
         if not op:
-            return TestCaseGenerationHandle(name=None)
+            return _TestCaseGenerationHandle(name=None)
 
         # progress
         if hasattr(op, "metadata") and op.metadata:
@@ -114,14 +109,8 @@ class TestCaseGenerationHandle:
         else:
             meta_dict = {}
 
-        return TestCaseGenerationHandle(
+        return _TestCaseGenerationHandle(
             name=op.name,
-            create_time=op.create_time,
-            creator=op.creator,
-            update_time=op.update_time,
-            updater=op.updater,
-            delete_time=op.delete_time,
-            deleter=op.deleter,
             progress=meta_dict.get("progress"),
             progress_message=meta_dict.get("progressMessage"),
             error=op.error,
@@ -193,6 +182,7 @@ class Test:
     create_time: Optional[datetime.datetime] = None
     update_time: Optional[datetime.datetime] = None
     _client: Optional[api.ApiClient] = None
+    _gen_tc_op_name: Optional[str] = None
 
     def __post_init__(self):
         if self._client:
@@ -271,7 +261,8 @@ class Test:
         model: Optional[str] = None,
         base_llm_model: Optional[str] = None,
         generators: Optional[List[TestCaseGenerator]] = None,
-    ) -> "TestCaseGenerationHandle":
+        existing_collection: Optional[str] = None,
+    ) -> None:
         """Generates test cases based on the documents of the Test.
 
         Args:
@@ -280,6 +271,9 @@ class Test:
             model (str): Model to use for generating the prompts.
             base_llm_model (str): Base LLM model to use for generating the prompts.
             generators (List[TestCaseGenerator]): Methods to use for generation.
+            existing_collection (str): ID or the resource name of the existing
+                collection, from which prompts will be generated.
+                NOTE: This option works only for the H2OGPTe model host ATM.
         """
 
         req = models.TestServiceGenerateTestCasesRequest(
@@ -287,61 +281,64 @@ class Test:
             model=model or None,
             base_llm_model=base_llm_model or None,
             generators=[g.to_api_proto() for g in generators] if generators else None,
+            h2ogpte_collection_id=existing_collection or None,
         )
 
         res = self._test_api.test_service_generate_test_cases(self.key, req)
 
-        return TestCaseGenerationHandle._from_operation(res)
+        op: models.V1Operation | None = res.operation
+        self._gen_tc_op_name = op.name if op else None
 
     def wait_for_test_case_generation(
-        self,
-        handle: TestCaseGenerationHandle,
-        timeout: Optional[float] = None,
-        verbose: bool = False,
-    ) -> TestCaseGenerationHandle:
+        self, timeout: Optional[float] = None, verbose: bool = False
+    ) -> None:
         """Waits for the test case generation to finish.
 
         Args:
-            handle (TestCaseGenerationHandle): Handle of the test case generation.
             timeout (float): The maximum time to wait in seconds.
             verbose (bool): If True, prints the status of the handle while waiting.
         """
-        if not handle.name:
-            raise ValueError("Test case generation handle is not valid.")
-        elif handle.done:
-            return handle
+        if not self._gen_tc_op_name:
+            raise ValueError(
+                "There is no ongoing test case generation - the operation name is not "
+                "set."
+            )
 
         if verbose:
-            print(f"Waiting for test case generation to finish ({handle.name}):")
+            print(
+                f"Waiting for test case generation to finish ({self._gen_tc_op_name}):"
+            )
         if self._client:
             # exponential backoff
             wait_time = 1.0
             wait_coef = 1.6
-            wait_max = 20.0
+            wait_max = 8.0
             wait_total = 0.0
             timeout = timeout or float(2 * 24 * 60 * 60)  # 2 days
+            progress_bar = utils.ProgressBar()
             while wait_total < timeout:
-                handle = TestCaseGenerationHandle._from_operation(
-                    self._operation_api.operation_service_get_operation(handle.name)
+                handle = _TestCaseGenerationHandle._from_operation(
+                    self._operation_api.operation_service_get_operation(
+                        self._gen_tc_op_name
+                    )
                 )
 
                 if verbose:
-                    if handle.progress or handle.progress_message:
-                        progress = (
-                            int(handle.progress * 100.0) if handle.progress else 0
-                        )
-                        msg = f"{progress:>2}% - '{handle.progress_message}'"
-                    else:
-                        msg = " 0% - 'Initializing'"
-                    print(f"  {msg}")
+                    progress_bar.update(handle.progress or 0, handle.progress_message)
 
                 if handle.done:
-                    return handle
+                    if handle.error:
+                        raise RuntimeError(
+                            f"Test case generation failed: {handle.error}"
+                        )
+                    return
 
                 wait_time *= wait_coef
                 time.sleep(min(wait_time, wait_max))
         else:
-            raise ValueError("Cannot establish connection to Eval Studio host.")
+            raise ValueError(
+                "Unable to establish a connection to the Eval Studio host."
+            )
 
         raise TimeoutError("Waiting timeout has been reached.")
 

eval_studio_client/utils.py

@@ -0,0 +1,26 @@
+from typing import Optional
+
+
+class ProgressBar:
+    def __init__(self):
+        self.progress = 0.0
+        self.progress_message = "Initializing"
+        self._progress_max = 1.0
+
+    def update(self, progress: float, message: Optional[str] = None):
+        try:
+            self.progress = float(str(progress))
+        except ValueError:
+            self.progress = 0.0
+
+        if message:
+            self.progress_message = message or ""
+
+        self.print()
+
+    def print(self):
+        print(" " * len(self.progress_message), end="\r")
+        p_progress = int(self.progress / self._progress_max * 100)
+        p_hashes = p_progress // 5
+        p_msg = f"  {p_progress:>3}% |{'#' * p_hashes:<20}| {self.progress_message}"
+        print(p_msg, end="\r")

{eval_studio_client-0.8.2.dist-info → eval_studio_client-1.0.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: eval-studio-client
-Version: 0.8.2
+Version: 1.0.0
 Project-URL: Source, https://github.com/h2oai/eval-studio/tree/main/client-py/src/
 Project-URL: Issues, https://github.com/h2oai/eval-studio/issues
 Author-email: "H2O.ai" <support@h2o.ai>

{eval_studio_client-0.8.2.dist-info → eval_studio_client-1.0.0.dist-info}/RECORD

@@ -1,16 +1,17 @@
 eval_studio_client/__about__.py,sha256=7TnXVu0lNAY4UdQ_2iwTlAENGdigMUVBy6UmtWGB6sQ,30
 eval_studio_client/__init__.py,sha256=v8lXY_l4j3lAbIfW21nZFeWZX0sl4nKHbB29h2qYVU8,207
 eval_studio_client/client.py,sha256=khRFtcFNZHAMe1bA7SyvoLOPHVZQ2XJOZ3UB3gX8EKs,3307
-eval_studio_client/dashboards.py,sha256=S35kude0FSn-v0t-H1N6aHhsNhlmIgF3duKR8TUfKes,7331
+eval_studio_client/dashboards.py,sha256=TBMiO4OvTnWYSVuj2-EBxSdKQtEAb_HXgc9gXtRnu-s,8381
 eval_studio_client/documents.py,sha256=fjsbHnqZnouu0stCf_p15RgoszkY4_gIsbX1hiw7Xv8,3076
 eval_studio_client/evaluators.py,sha256=blJlWMswIGr1u6TQDiiO-fInYVnkBT0Y02J57o8Z094,2100
 eval_studio_client/insights.py,sha256=bhe6XBVJ61-2bcDdNe6HiZsu0sly8LeoYAKo1GkgK08,1199
-eval_studio_client/leaderboards.py,sha256=UZItYErAGRXDsae61iMnHXXjoAUFSPL-HTQ_eQnkIJI,7746
-eval_studio_client/models.py,sha256=4OFASuJF1OvIdVODqUk4Uv70cojIJ9CFz3U1nmPFJwI,19137
+eval_studio_client/leaderboards.py,sha256=NHko_kuPIXnbBdEDMK1MHQmHJRCHA7_Q1wx4eqBvBF8,8035
+eval_studio_client/models.py,sha256=nW1Wk6L89iWSjhMVk_sKmxSomKX3b6ANALbwWvbJ7Uk,21346
 eval_studio_client/perturbators.py,sha256=CtcWqEgPGpOcDHvYAQBlNDKnS-ZDBkL7Y_Ygsgpvikw,3133
 eval_studio_client/problems.py,sha256=rdGIfo7AqyxGhWMpbIDX1WXFoQvzKktKAWDKRde5VbY,1515
 eval_studio_client/test_labs.py,sha256=IEY98Ocu7WQcxZN_jy5YthVBoHAgHjgA2T93U7q0eYE,11260
-eval_studio_client/tests.py,sha256=n14-zM2J9oUKgKZQm2xjtg7f8MWxnL2Ov00jQqMP8fw,22512
+eval_studio_client/tests.py,sha256=_Qu6X4FoocYJ-liClXLQqIR91P7GjWmxpeyDhRl5JXI,22393
+eval_studio_client/utils.py,sha256=e5bsQVgNHYNSqSOthxlmncerPdgbvWwQaY_C-libuXk,764
 eval_studio_client/api/__init__.py,sha256=Ef5qooH4SLfYUqVBJl79oRKWYnXryDPZV4IXGfvG1Wc,15269
 eval_studio_client/api/api_client.py,sha256=yFQKmCsVhswcTbdGY4lf-61mf8FVm3Kfon8Qhe1sPKw,26431
 eval_studio_client/api/api_response.py,sha256=eMxw1mpmJcoGZ3gs9z6jM4oYoZ10Gjk333s9sKxGv7s,652
@@ -480,6 +481,6 @@ eval_studio_client/api/test/test_v1_update_test_response.py,sha256=pqTwL9SgoOM9k
 eval_studio_client/api/test/test_v1_who_am_i_response.py,sha256=bNbjL5-b-4asyziW6znJhuU2yrzd9RgJa2ZiNw3e6YA,1523
 eval_studio_client/api/test/test_who_am_i_service_api.py,sha256=gYWKFamJMyVne2QaOSPz6WEkxExRuAphMGKf1nFayLU,898
 eval_studio_client/gen/openapiv2/eval_studio.swagger.json,sha256=2jOBBxQ2H2mS9C_nlqoTrTiYMmCLaUFQym6su3fXJ8I,210976
-eval_studio_client-0.8.2.dist-info/METADATA,sha256=hza1__A-Rky7RO8E8KyQmkb-KXvODW1wNLZPAWCJWBk,707
-eval_studio_client-0.8.2.dist-info/WHEEL,sha256=C2FUgwZgiLbznR-k0b_5k3Ai_1aASOXDss3lzCUsUug,87
-eval_studio_client-0.8.2.dist-info/RECORD,,
+eval_studio_client-1.0.0.dist-info/METADATA,sha256=l8XLUMIu-W4pHRG8fs1IZek_bGIEiFtDRyPjPGkpQrY,707
+eval_studio_client-1.0.0.dist-info/WHEEL,sha256=C2FUgwZgiLbznR-k0b_5k3Ai_1aASOXDss3lzCUsUug,87
+eval_studio_client-1.0.0.dist-info/RECORD,,