deepeval 3.7.0__tar.gz → 3.7.2__tar.gz

{deepeval-3.7.0 → deepeval-3.7.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deepeval
-Version: 3.7.0
+Version: 3.7.2
 Summary: The LLM Evaluation Framework
 Home-page: https://github.com/confident-ai/deepeval
 License: Apache-2.0
@@ -32,7 +32,7 @@ Requires-Dist: pyfiglet
 Requires-Dist: pytest
 Requires-Dist: pytest-asyncio
 Requires-Dist: pytest-repeat
-Requires-Dist: pytest-rerunfailures (>=12.0,<13.0)
+Requires-Dist: pytest-rerunfailures
 Requires-Dist: pytest-xdist
 Requires-Dist: python-dotenv (>=1.1.1,<2.0.0)
 Requires-Dist: requests (>=2.31.0,<3.0.0)
@@ -439,6 +439,7 @@ Using `.env.local` or `.env` is optional. If they are missing, DeepEval uses you
 ```bash
 cp .env.example .env.local
 # then edit .env.local (ignored by git)
+```
 
 <br />
 

{deepeval-3.7.0 → deepeval-3.7.2}/README.md

@@ -389,6 +389,7 @@ Using `.env.local` or `.env` is optional. If they are missing, DeepEval uses you
 ```bash
 cp .env.example .env.local
 # then edit .env.local (ignored by git)
+```
 
 <br />
 

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/__init__.py

@@ -102,9 +102,5 @@ def update_warning_opt_in():
     return os.getenv("DEEPEVAL_UPDATE_WARNING_OPT_IN") == "1"
 
 
-def is_read_only_env():
-    return os.getenv("DEEPEVAL_FILE_SYSTEM") == "READ_ONLY"
-
-
 if update_warning_opt_in():
     check_for_update()

deepeval-3.7.2/deepeval/_version.py

@@ -0,0 +1 @@
+__version__: str = "3.7.2"

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/cli/main.py

@@ -1484,6 +1484,11 @@ def set_gemini_model_env(
     google_cloud_location: Optional[str] = typer.Option(
         None, "--location", help="Google Cloud location"
     ),
+    google_service_account_key: Optional[str] = typer.Option(
+        None,
+        "--service-account-key",
+        help="Google Service Account Key for Gemini",
+    ),
     save: Optional[str] = typer.Option(
         None,
         "--save",
@@ -1513,6 +1518,8 @@ def set_gemini_model_env(
             settings.GOOGLE_CLOUD_PROJECT = google_cloud_project
         if google_cloud_location:
             settings.GOOGLE_CLOUD_LOCATION = google_cloud_location
+        if google_service_account_key:
+            settings.GOOGLE_SERVICE_ACCOUNT_KEY = google_service_account_key
         if model_name:
             settings.GEMINI_MODEL_NAME = model_name
 

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/confident/api.py

@@ -27,6 +27,10 @@ retryable_exceptions = requests.exceptions.SSLError
 
 
 def get_base_api_url():
+    s = get_settings()
+    if s.CONFIDENT_BASE_URL:
+        base_url = s.CONFIDENT_BASE_URL.rstrip("/")
+        return base_url
     region = KEY_FILE_HANDLER.fetch_data(KeyValues.CONFIDENT_REGION)
     if region == "EU":
         return API_BASE_URL_EU
@@ -87,6 +91,7 @@ class Endpoints(Enum):
     DATASET_ALIAS_QUEUE_ENDPOINT = "/v1/datasets/:alias/queue"
 
     TEST_RUN_ENDPOINT = "/v1/test-run"
+    EXPERIMENT_ENDPOINT = "/v1/experiment"
     METRIC_DATA_ENDPOINT = "/v1/metric-data"
     TRACES_ENDPOINT = "/v1/traces"
     ANNOTATIONS_ENDPOINT = "/v1/annotations"
@@ -115,7 +120,7 @@ class Api:
         self.api_key = api_key
         self._headers = {
             "Content-Type": "application/json",
-            "CONFIDENT_API_KEY": api_key,
+            "CONFIDENT-API-KEY": api_key,
             "X-DeepEval-Version": deepeval.__version__,
         }
         self.base_api_url = get_base_api_url()

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/config/settings.py

@@ -229,6 +229,11 @@ class Settings(BaseSettings):
     API_KEY: Optional[SecretStr] = None
     CONFIDENT_API_KEY: Optional[SecretStr] = None
 
+    # ======
+    # Base URL for Confident AI API server
+    # ======
+    CONFIDENT_BASE_URL: Optional[str] = None
+
     # General
     TEMPERATURE: Optional[confloat(ge=0, le=2)] = None
 

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/evaluate/compare.py

@@ -1,5 +1,6 @@
 from typing import Optional, List, Dict, Callable
 import asyncio
+import time
 from rich.progress import (
     Progress,
     TextColumn,
@@ -8,24 +9,74 @@ from rich.progress import (
     TaskProgressColumn,
 )
 from collections import Counter
+import json
 
 from deepeval.errors import MissingTestCaseParamsError
 from deepeval.evaluate.configs import AsyncConfig, DisplayConfig, ErrorConfig
-from deepeval.test_case import ArenaTestCase
+from deepeval.test_case import ArenaTestCase, Contestant
+from deepeval.test_case.api import create_api_test_case
 from deepeval.metrics import ArenaGEval
-from deepeval.utils import add_pbar, update_pbar, custom_console
-from deepeval.utils import get_or_create_event_loop
+from deepeval.utils import (
+    add_pbar,
+    update_pbar,
+    custom_console,
+    get_or_create_event_loop,
+    open_browser,
+)
+from deepeval.test_run.test_run import (
+    TestRun,
+    MetricData,
+    TestRunEncoder,
+    MetricScores,
+    console,
+)
+from deepeval.test_run.hyperparameters import (
+    process_hyperparameters,
+)
+from deepeval.confident.api import Api, Endpoints, HttpMethods, is_confident
 from deepeval.telemetry import capture_evaluation_run
+from deepeval.test_run.api import LLMApiTestCase
+from deepeval.evaluate.utils import create_arena_metric_data
+from deepeval.evaluate.types import PostExperimentRequest
 
 
 def compare(
     test_cases: List[ArenaTestCase],
     metric: ArenaGEval,
+    name: str = "compare()",
     # Configs
     async_config: Optional[AsyncConfig] = AsyncConfig(),
     display_config: Optional[DisplayConfig] = DisplayConfig(),
     error_config: Optional[ErrorConfig] = ErrorConfig(),
 ) -> Dict[str, int]:
+
+    # Prepare test run map
+    unique_contestant_names = set(
+        [
+            contestant.name
+            for test_case in test_cases
+            for contestant in test_case.contestants
+        ]
+    )
+    test_run_map: Dict[str, TestRun] = {}
+    for contestant_name in unique_contestant_names:
+        test_run = TestRun(
+            identifier=contestant_name,
+            test_passed=0,
+            test_failed=0,
+        )
+        test_run.metrics_scores = [
+            MetricScores(
+                metric=metric.name,
+                scores=[],
+                passes=0,
+                fails=0,
+                errors=0,
+            )
+        ]
+        test_run_map[contestant_name] = test_run
+
+    start_time = time.time()
     with capture_evaluation_run("compare()"):
         if async_config.run_async:
             loop = get_or_create_event_loop()
@@ -39,6 +90,7 @@ def compare(
                     throttle_value=async_config.throttle_value,
                     max_concurrent=async_config.max_concurrent,
                     skip_on_missing_params=error_config.skip_on_missing_params,
+                    test_run_map=test_run_map,
                 )
             )
         else:
@@ -49,7 +101,10 @@ def compare(
                 verbose_mode=display_config.verbose_mode,
                 show_indicator=display_config.show_indicator,
                 skip_on_missing_params=error_config.skip_on_missing_params,
+                test_run_map=test_run_map,
             )
+    end_time = time.time()
+    run_duration = end_time - start_time
 
     # Aggregate winners
     winner_counts = Counter()
@@ -57,7 +112,13 @@ def compare(
         if winner:
             winner_counts[winner] += 1
 
-    print(winner_counts)
+    process_test_runs(test_run_map=test_run_map, test_cases=test_cases)
+    wrap_up_experiment(
+        name=name,
+        test_runs=list(test_run_map.values()),
+        winner_counts=winner_counts,
+        run_duration=run_duration,
+    )
     return dict(winner_counts)
 
 
@@ -70,6 +131,7 @@ async def a_execute_arena_test_cases(
     throttle_value: int,
     skip_on_missing_params: bool,
     max_concurrent: int,
+    test_run_map: Dict[str, TestRun],
 ) -> List[str]:
     semaphore = asyncio.Semaphore(max_concurrent)
 
@@ -104,6 +166,8 @@ async def a_execute_arena_test_cases(
                 else metric.verbose_mode
             ),
         )
+
+        start_time = time.perf_counter()
         winner = await _a_handle_metric_measurement(
             metric=metric_copy,
             test_case=test_case,
@@ -112,10 +176,21 @@ async def a_execute_arena_test_cases(
             _progress=progress,
             _pbar_id=pbar_test_case_id,
         )
+        end_time = time.perf_counter()
+        run_duration = end_time - start_time
+
         if winner:
             winners.append(winner)
 
         update_pbar(progress, pbar_id)
+        update_test_run_map(
+            test_case=test_case,
+            index=index,
+            test_run_map=test_run_map,
+            metric_copy=metric_copy,
+            winner=winner,
+            run_duration=run_duration,
+        )
 
     # Create tasks for all test cases
     if show_indicator:
@@ -156,6 +231,7 @@ def execute_arena_test_cases(
     skip_on_missing_params: bool,
     show_indicator: bool,
     verbose_mode: Optional[bool] = None,
+    test_run_map: Optional[Dict[str, TestRun]] = None,
 ) -> List[str]:
     """
     Non-async version of comparing arena test cases.
@@ -183,6 +259,8 @@ def execute_arena_test_cases(
                     else metric.verbose_mode
                 ),
             )
+
+            start_time = time.perf_counter()
             winner = _handle_metric_measurement(
                 metric=metric_copy,
                 test_case=test_case,
@@ -191,10 +269,21 @@ def execute_arena_test_cases(
                 _progress=progress,
                 _pbar_id=pbar_test_case_id,
             )
+            end_time = time.perf_counter()
+            run_duration = end_time - start_time
+
             if winner:
                 winners.append(winner)
 
             update_pbar(progress, pbar_id)
+            update_test_run_map(
+                test_case=test_case,
+                index=i,
+                test_run_map=test_run_map,
+                metric_copy=metric_copy,
+                winner=winner,
+                run_duration=run_duration,
+            )
 
     if show_indicator:
         progress = Progress(
@@ -313,3 +402,129 @@ async def _a_handle_metric_measurement(
                 return None
             else:
                 raise
+
+
+def update_test_run_map(
+    test_case: ArenaTestCase,
+    index: int,
+    test_run_map: Dict[str, TestRun],
+    metric_copy: ArenaGEval,
+    winner: str,
+    run_duration: float,
+):
+    for contestant in test_case.contestants:
+        test_run = test_run_map.get(contestant.name)
+
+        # update test cases in test run
+        api_test_case: LLMApiTestCase = create_api_test_case(
+            test_case=contestant.test_case, index=index
+        )
+        metric_data: MetricData = create_arena_metric_data(
+            metric_copy, contestant.name
+        )
+        api_test_case.update_metric_data(metric_data)
+        api_test_case.update_run_duration(run_duration)
+        test_run.add_test_case(api_test_case)
+
+        # update other test run attributes
+        if test_run.run_duration is None:
+            test_run.run_duration = 0.0
+        test_run.run_duration += run_duration
+
+        # Ensure test_passed and test_failed are initialized
+        if test_run.test_passed is None:
+            test_run.test_passed = 0
+        if test_run.test_failed is None:
+            test_run.test_failed = 0
+
+        if winner == contestant:
+            test_run.test_passed += 1
+        else:
+            test_run.test_failed += 1
+
+        # update metric scores
+        test_run.metrics_scores[0].metric = metric_copy.name
+        test_run.metrics_scores[0].scores.append(
+            1 if winner == contestant else 0
+        )
+        test_run.metrics_scores[0].passes += 1 if winner == contestant else 0
+        test_run.metrics_scores[0].fails += 1 if winner != contestant else 0
+        test_run.metrics_scores[0].errors += 0
+
+
+def process_test_runs(
+    test_run_map: Dict[str, TestRun],
+    test_cases: List[ArenaTestCase],
+):
+    hyperparameters_map = {
+        contestant_name: {} for contestant_name in test_run_map.keys()
+    }
+
+    for test_case in test_cases:
+        for contestant in test_case.contestants:
+            if contestant.hyperparameters:
+                hyperparameters_map[contestant.name].update(
+                    contestant.hyperparameters
+                )
+
+    for contestant_name, hyperparameters in hyperparameters_map.items():
+        test_run = test_run_map.get(contestant_name)
+        test_run.hyperparameters = process_hyperparameters(hyperparameters)
+
+
+def wrap_up_experiment(
+    name: str,
+    test_runs: List[TestRun],
+    winner_counts: Counter,
+    run_duration: float,
+):
+    winner_breakdown = []
+    for contestant, wins in winner_counts.most_common():
+        winner_breakdown.append(
+            f"    » [bold green]{contestant}[/bold green]: {wins} wins"
+        )
+    winner_text = (
+        "\n".join(winner_breakdown) if winner_breakdown else "No winners"
+    )
+    console.print(
+        f"\n🎉 Arena completed! (time taken: {round(run_duration, 2)}s | token cost: {test_runs[0].evaluation_cost if test_runs else 0} USD)\n"
+        f"🏆 Results ({sum(winner_counts.values())} total test cases):\n"
+        f"{winner_text}\n\n"
+    )
+
+    if not is_confident():
+        console.print(
+            f"{'=' * 80}\n"
+            f"\n» Want to share experiments with your team? ❤️ 🏟️\n"
+            f"  » Run [bold]'deepeval login'[/bold] to analyze and save arena results on [rgb(106,0,255)]Confident AI[/rgb(106,0,255)].\n\n"
+        )
+        return
+
+    try:
+        api = Api()
+        experiment_request = PostExperimentRequest(
+            testRuns=test_runs, name=name
+        )
+
+        try:
+            body = experiment_request.model_dump(
+                by_alias=True, exclude_none=True
+            )
+        except AttributeError:
+            body = experiment_request.dict(by_alias=True, exclude_none=True)
+        json_str = json.dumps(body, cls=TestRunEncoder)
+        body = json.loads(json_str)
+
+        _, link = api.send_request(
+            method=HttpMethods.POST,
+            endpoint=Endpoints.EXPERIMENT_ENDPOINT,
+            body=body,
+        )
+        console.print(
+            "[rgb(5,245,141)]✓[/rgb(5,245,141)] Done 🎉! View results on "
+            f"[link={link}]{link}[/link]"
+        )
+        open_browser(link)
+
+    except Exception:
+        raise

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/evaluate/types.py

@@ -4,6 +4,7 @@ from pydantic import BaseModel
 
 from deepeval.test_run.api import MetricData, TurnApi
 from deepeval.test_case import MLLMImage
+from deepeval.test_run import TestRun
 
 
 @dataclass
@@ -29,3 +30,8 @@ class EvaluationResult(BaseModel):
     test_results: List[TestResult]
     confident_link: Optional[str]
     test_run_id: Optional[str]
+
+
+class PostExperimentRequest(BaseModel):
+    testRuns: List[TestRun]
+    name: Optional[str]

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/evaluate/utils.py

@@ -8,6 +8,7 @@ from deepeval.utils import format_turn
 from deepeval.test_run.test_run import TestRunResultDisplay
 from deepeval.dataset import Golden
 from deepeval.metrics import (
+    ArenaGEval,
     BaseMetric,
     BaseConversationalMetric,
     BaseMultimodalMetric,
@@ -84,6 +85,35 @@ def create_metric_data(metric: BaseMetric) -> MetricData:
         )
 
 
+def create_arena_metric_data(metric: ArenaGEval, contestant: str) -> MetricData:
+    if metric.error is not None:
+        return MetricData(
+            name=metric.__name__,
+            threshold=1,
+            score=None,
+            reason=None,
+            success=False,
+            strictMode=True,
+            evaluationModel=metric.evaluation_model,
+            error=metric.error,
+            evaluationCost=metric.evaluation_cost,
+            verboseLogs=metric.verbose_logs,
+        )
+    else:
+        return MetricData(
+            name=metric.__name__,
+            score=1 if contestant == metric.winner else 0,
+            threshold=1,
+            reason=metric.reason,
+            success=metric.is_successful(),
+            strictMode=True,
+            evaluationModel=metric.evaluation_model,
+            error=None,
+            evaluationCost=metric.evaluation_cost,
+            verboseLogs=metric.verbose_logs,
+        )
+
+
 def create_test_result(
     api_test_case: Union[LLMApiTestCase, ConversationalApiTestCase],
 ) -> TestResult:

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/key_handler.py

@@ -70,6 +70,7 @@ class ModelKeyValues(Enum):
     GOOGLE_GENAI_USE_VERTEXAI = "GOOGLE_GENAI_USE_VERTEXAI"
     GOOGLE_CLOUD_PROJECT = "GOOGLE_CLOUD_PROJECT"
     GOOGLE_CLOUD_LOCATION = "GOOGLE_CLOUD_LOCATION"
+    GOOGLE_SERVICE_ACCOUNT_KEY = "GOOGLE_SERVICE_ACCOUNT_KEY"
     # LiteLLM
     USE_LITELLM = "USE_LITELLM"
     LITELLM_MODEL_NAME = "LITELLM_MODEL_NAME"

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/arena_g_eval/arena_g_eval.py

@@ -46,7 +46,11 @@ class ArenaGEval(BaseArenaMetric):
         self.criteria = criteria
         self.model, self.using_native_model = initialize_model(model)
         self.evaluation_model = self.model.get_model_name()
-        self.evaluation_steps = evaluation_steps
+        self.evaluation_steps = (
+            evaluation_steps
+            if evaluation_steps and len(evaluation_steps) > 0
+            else None
+        )
         self.async_mode = async_mode
         self.verbose_mode = verbose_mode
         self._include_g_eval_suffix = _include_g_eval_suffix

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/arena_g_eval/utils.py

@@ -89,10 +89,10 @@ class FormattedArenaTestCase:
 def format_arena_test_case(
     evaluation_params: List[LLMTestCaseParams], test_case: ArenaTestCase
 ) -> Tuple[FormattedArenaTestCase, Dict[str, str]]:
-    case = next(iter(test_case.contestants.values()))
+    case = next(iter([case.test_case for case in test_case.contestants]))
 
     # Create dummy name mapping
-    real_names = list(test_case.contestants.keys())
+    real_names = list([case.name for case in test_case.contestants])
     available_fake_names = FAKE_NAMES.copy()
     random.shuffle(available_fake_names)
 
@@ -119,10 +119,10 @@ def format_arena_test_case(
             else None
         ),
         contestants={
-            contestant: construct_formatted_llm_test_case(
-                evaluation_params, test_case
+            contestant.name: construct_formatted_llm_test_case(
+                evaluation_params, contestant.test_case
             )
-            for contestant, test_case in test_case.contestants.items()
+            for contestant in test_case.contestants
         },
         dummy_to_real_names=dummy_to_real_names,
     )

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/conversational_g_eval/conversational_g_eval.py

@@ -9,6 +9,8 @@ from deepeval.metrics.g_eval.utils import (
     construct_conversational_g_eval_turn_params_string,
     construct_non_turns_test_case_string,
     format_rubrics,
+    validate_and_sort_rubrics,
+    validate_criteria_and_evaluation_steps,
 )
 from deepeval.test_case import (
     TurnParams,
@@ -63,27 +65,16 @@ class ConversationalGEval(BaseConversationalMetric):
 
         self.evaluation_params = evaluation_params
 
-        # Check if both criteria and evaluation_steps are not None at the same time
-        if criteria is None and evaluation_steps is None:
-            raise ValueError(
-                "Either 'criteria' or 'evaluation_steps' must be provided."
-            )
-
-        # Check if criteria is provided, it cannot be an empty string
-        if criteria is not None and not criteria.strip():
-            raise ValueError("Criteria provided cannot be an empty string.")
-
-        # Check if evaluation_steps is provided, it cannot be an empty list
-        if evaluation_steps is not None and len(evaluation_steps) == 0:
-            raise ValueError(
-                "'evaluation_steps' must not be an empty list. Either omit evaluation steps or include a non-empty list of steps."
-            )
-
+        validate_criteria_and_evaluation_steps(criteria, evaluation_steps)
         self.criteria = criteria
-        self.rubric = rubric
+        self.rubric = validate_and_sort_rubrics(rubric)
         self.model, self.using_native_model = initialize_model(model)
         self.evaluation_model = self.model.get_model_name()
-        self.evaluation_steps = evaluation_steps
+        self.evaluation_steps = (
+            evaluation_steps
+            if evaluation_steps and len(evaluation_steps) > 0
+            else None
+        )
         self.threshold = 1 if strict_mode else threshold
         self.strict_mode = strict_mode
         self.async_mode = async_mode

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/g_eval/g_eval.py

@@ -61,7 +61,11 @@ class GEval(BaseMetric):
         self.score_range_span = self.score_range[1] - self.score_range[0]
         self.model, self.using_native_model = initialize_model(model)
         self.evaluation_model = self.model.get_model_name()
-        self.evaluation_steps = evaluation_steps
+        self.evaluation_steps = (
+            evaluation_steps
+            if evaluation_steps and len(evaluation_steps) > 0
+            else None
+        )
         self.threshold = 1 if strict_mode else threshold
         self.top_logprobs = top_logprobs
         self.strict_mode = strict_mode

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/g_eval/utils.py

@@ -77,7 +77,7 @@ def validate_criteria_and_evaluation_steps(
 def validate_and_sort_rubrics(
     rubrics: Optional[List[Rubric]] = None,
 ) -> Optional[List[Rubric]]:
-    if rubrics is None:
+    if rubrics is None or len(rubrics) == 0:
         return None
 
     # Sort rubrics by start of range

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/multimodal_metrics/multimodal_g_eval/multimodal_g_eval.py

@@ -64,7 +64,11 @@ class MultimodalGEval(BaseMultimodalMetric):
         self.rubric = validate_and_sort_rubrics(rubric)
         self.model, self.using_native_model = initialize_multimodal_model(model)
         self.evaluation_model = self.model.get_model_name()
-        self.evaluation_steps = evaluation_steps
+        self.evaluation_steps = (
+            evaluation_steps
+            if evaluation_steps and len(evaluation_steps) > 0
+            else None
+        )
         self.threshold = 1 if strict_mode else threshold
         self.top_logprobs = top_logprobs
         self.strict_mode = strict_mode

{deepeval-3.7.0 → deepeval-3.7.2}/deepeval/metrics/utils.py

@@ -270,7 +270,7 @@ def check_arena_test_case_params(
             f"Expected ArenaTestCase, got {type(arena_test_case).__name__}"
         )
 
-    cases = list(arena_test_case.contestants.values())
+    cases = [contestant.test_case for contestant in arena_test_case.contestants]
     ref_input = cases[0].input
     for case in cases[1:]:
         if case.input != ref_input: