opik-optimizer 2.1.2__py3-none-any.whl → 2.2.0__py3-none-any.whl

opik_optimizer/reporting_utils.py

@@ -14,6 +14,24 @@ from .utils import get_optimization_run_url_by_id
 PANEL_WIDTH = 70
 
 
+def safe_percentage_change(current: float, baseline: float) -> tuple[float, bool]:
+    """
+    Calculate percentage change safely, handling division by zero.
+
+    Args:
+        current: Current value
+        baseline: Baseline value to compare against
+
+    Returns:
+        Tuple of (percentage_change, has_percentage) where:
+        - percentage_change: The percentage change if calculable, otherwise 0
+        - has_percentage: True if percentage was calculated, False if baseline was zero
+    """
+    if baseline == 0:
+        return 0.0, False
+    return ((current - baseline) / baseline) * 100, True
+
+
 def get_console(*args: Any, **kwargs: Any) -> Console:
     console = Console(*args, **kwargs)
     console.is_jupyter = False
@@ -36,35 +54,44 @@ def convert_tqdm_to_rich(description: str | None = None, verbose: int = 1) -> An
 
     from opik.evaluation import report
 
+    # Store original functions
+    original_display_experiment_results = report.display_experiment_results
+    original_display_experiment_link = report.display_experiment_link
+
+    # Replace with no-ops
     report.display_experiment_results = lambda *args, **kwargs: None
     report.display_experiment_link = lambda *args, **kwargs: None
 
     try:
         yield
     finally:
+        # Restore everything
         opik.evaluation.engine.evaluation_tasks_executor._tqdm = original__tqdm
+        report.display_experiment_results = original_display_experiment_results
+        report.display_experiment_link = original_display_experiment_link
 
 
 @contextmanager
 def suppress_opik_logs() -> Any:
     """Suppress Opik startup logs by temporarily increasing the log level."""
-    # Optimizer log level
-    optimizer_logger = logging.getLogger("opik_optimizer")
+    # Get all loggers we need to suppress
+    opik_client_logger = logging.getLogger("opik.api_objects.opik_client")
+    opik_logger = logging.getLogger("opik")
 
-    # Get the Opik logger
-    opik_logger = logging.getLogger("opik.api_objects.opik_client")
+    # Store original log levels
+    original_client_level = opik_client_logger.level
+    original_opik_level = opik_logger.level
 
-    # Store original log level
-    original_level = opik_logger.level
-
-    # Set log level to ERROR to suppress INFO messages
-    opik_logger.setLevel(optimizer_logger.level)
+    # Set log level to WARNING to suppress INFO messages
+    opik_client_logger.setLevel(logging.WARNING)
+    opik_logger.setLevel(logging.WARNING)
 
     try:
         yield
     finally:
-        # Restore original log level
-        opik_logger.setLevel(original_level)
+        # Restore original log levels
+        opik_client_logger.setLevel(original_client_level)
+        opik_logger.setLevel(original_opik_level)
 
 
 def display_messages(messages: list[dict[str, str]], prefix: str = "") -> None:
@@ -196,18 +223,18 @@ def display_result(
     content: Text | Panel = []
 
     if best_score > initial_score:
-        if initial_score == 0:
+        perc_change, has_percentage = safe_percentage_change(best_score, initial_score)
+        if has_percentage:
             content += [
                 Text(
-                    f"Prompt was optimized and improved from {initial_score:.4f} to {best_score:.4f}",
+                    f"Prompt was optimized and improved from {initial_score:.4f} to {best_score:.4f} ({perc_change:.2%})",
                     style="bold green",
                 )
             ]
         else:
-            perc_change = (best_score - initial_score) / initial_score
             content += [
                 Text(
-                    f"Prompt was optimized and improved from {initial_score:.4f} to {best_score:.4f} ({perc_change:.2%})",
+                    f"Prompt was optimized and improved from {initial_score:.4f} to {best_score:.4f}",
                     style="bold green",
                 )
             ]

opik_optimizer/task_evaluator.py

@@ -5,6 +5,7 @@ from collections.abc import Callable
 import opik
 from opik.evaluation import evaluator as opik_evaluator
 from opik.evaluation.metrics import base_metric, score_result
+from . import multi_metric_objective
 
 logger = logging.getLogger(__name__)
 
@@ -14,9 +15,20 @@ def _create_metric_class(metric: Callable) -> base_metric.BaseMetric:
         def __init__(self) -> None:
             self.name = metric.__name__
 
-        def score(self, llm_output: str, **kwargs: Any) -> score_result.ScoreResult:
+        def score(
+            self, llm_output: str, **kwargs: Any
+        ) -> score_result.ScoreResult | list[score_result.ScoreResult]:
             try:
                 metric_val = metric(dataset_item=kwargs, llm_output=llm_output)
+
+                if isinstance(metric, multi_metric_objective.MultiMetricObjective):
+                    if (
+                        hasattr(metric_val, "metadata")
+                        and "raw_score_results" in metric_val.metadata
+                    ):
+                        return [metric_val, *metric_val.metadata["raw_score_results"]]
+                    else:
+                        return [metric_val]
                 if isinstance(metric_val, score_result.ScoreResult):
                     return score_result.ScoreResult(
                         name=self.name,
@@ -107,15 +119,20 @@ def evaluate(
     if not result.test_results:
         return 0.0
 
-    # We may allow score aggregation customization.
-    score_results: list[score_result.ScoreResult] = [
-        test_result.score_results[0] for test_result in result.test_results
-    ]
-    if not score_results:
+    # Filter score results to only include the objective metric
+    objective_metric_name = metric.__name__
+    objective_score_results: list[score_result.ScoreResult] = []
+    for test_result in result.test_results:
+        for score_result_ in test_result.score_results:
+            if score_result_.name == objective_metric_name:
+                objective_score_results.append(score_result_)
+                break
+
+    if not objective_score_results:
         return 0.0
 
-    avg_score = sum([score_result_.value for score_result_ in score_results]) / len(
-        score_results
-    )
+    avg_score = sum(
+        [score_result_.value for score_result_ in objective_score_results]
+    ) / len(objective_score_results)
 
     return avg_score

opik_optimizer/utils/core.py

@@ -327,9 +327,16 @@ def create_litellm_agent_class(
         class LiteLLMAgent(OptimizableAgent):
             model = prompt.model
             model_kwargs = prompt.model_kwargs
-            project_name = prompt.project_name
             optimizer = optimizer_ref
 
+            def __init__(
+                self, prompt: "ChatPrompt", project_name: str | None = None
+            ) -> None:
+                # Get project_name from optimizer if available
+                if project_name is None and hasattr(self.optimizer, "project_name"):
+                    project_name = self.optimizer.project_name
+                super().__init__(prompt, project_name=project_name)
+
             def invoke(
                 self, messages: list[dict[str, str]], seed: int | None = None
             ) -> str:
@@ -342,9 +349,16 @@ def create_litellm_agent_class(
         class LiteLLMAgent(OptimizableAgent):  # type: ignore[no-redef]
             model = prompt.model
             model_kwargs = prompt.model_kwargs
-            project_name = prompt.project_name
             optimizer = optimizer_ref
 
+            def __init__(
+                self, prompt: "ChatPrompt", project_name: str | None = None
+            ) -> None:
+                # Get project_name from optimizer if available
+                if project_name is None and hasattr(self.optimizer, "project_name"):
+                    project_name = self.optimizer.project_name
+                super().__init__(prompt, project_name=project_name)
+
     return LiteLLMAgent
 
 

opik_optimizer/utils/prompt_segments.py

@@ -175,8 +175,7 @@ def apply_segment_updates(
         function_map=prompt.function_map,
         model=prompt.model,
         invoke=prompt.invoke,
-        project_name=prompt.project_name,
-        **prompt.model_kwargs,
+        model_parameters=prompt.model_kwargs,
     )
 
 

{opik_optimizer-2.1.2.dist-info → opik_optimizer-2.2.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: opik_optimizer
-Version: 2.1.2
+Version: 2.2.0
 Summary: Agent optimization with Opik
 Home-page: https://github.com/comet-ml/opik
 Author: Comet ML
@@ -8,13 +8,12 @@ Author-email: Comet ML <support@comet.com>
 License: Apache 2.0
 Project-URL: Homepage, https://github.com/comet-ml/opik/blob/main/sdks/opik_optimizer/README.md
 Project-URL: Repository, https://github.com/comet-ml/opik
-Requires-Python: >=3.10,<3.13
+Requires-Python: >=3.10,<3.14
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: datasets
 Requires-Dist: deap>=1.4.3
 Requires-Dist: diskcache
-Requires-Dist: dspy<3
 Requires-Dist: gepa>=0.0.7
 Requires-Dist: ujson
 Requires-Dist: hf_xet

{opik_optimizer-2.1.2.dist-info → opik_optimizer-2.2.0.dist-info}/RECORD

@@ -1,13 +1,14 @@
-opik_optimizer/__init__.py,sha256=VwryQ5bSOmJSl4CiacCIv_UF_In8Zho54fQ3FUR8pyk,1573
+opik_optimizer/__init__.py,sha256=HsEIWyxeUJhzCvuML5SjBHFWtm-b5LSHyE9GRYytyeI,1592
 opik_optimizer/_throttle.py,sha256=1JXIhYlo0IaqCgwmNB0Hnh9CYhYPkwRFdVGIcE7pVNg,1362
-opik_optimizer/base_optimizer.py,sha256=XryBkUTs4FQmHcBtVm63EJIKWrTvwqduUZ6ArHzYQko,21520
+opik_optimizer/base_optimizer.py,sha256=VpH6JSalcoewGkIN0h77_crCAkx5ffQtjNhaD0Xtazg,28350
 opik_optimizer/cache_config.py,sha256=Xd3NdUsL7bLQWoNe3pESqH4nHucU1iNTSGp-RqbwDog,599
 opik_optimizer/logging_config.py,sha256=TmxX0C1P20amxoXuiNQvlENOjdSNfWwvL8jFy206VWM,3837
-opik_optimizer/optimizable_agent.py,sha256=R0_BdwdHyZGWTw3oSvTg8FULDOYM8XaTiPNR3qV8DkQ,6344
+opik_optimizer/multi_metric_objective.py,sha256=y4jqirnhkfhB7SWonI4ldYg5fWG4JGfAxqu7ylRD1J4,1178
+opik_optimizer/optimizable_agent.py,sha256=gB1ALuVPyEmXOTVYeK2i-inBAO-6JMZSjOrmj37okgQ,6514
 opik_optimizer/optimization_result.py,sha256=sG-Yr-hOaH9zx_I5S6_W3v6j8nPUhwYdS333jVM4Gus,17218
 opik_optimizer/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-opik_optimizer/reporting_utils.py,sha256=dcECFmzZ_J-DKoukMDEE_fm7X8sdQyl_ijTddvQtepE,8287
-opik_optimizer/task_evaluator.py,sha256=1hILYwJLtn7XpPX96JjubnlMasmudVTHMVK3pmd22bE,4312
+opik_optimizer/reporting_utils.py,sha256=Gx69W16FfIpavH_o0WvnGJPIpqHAjJm4GNHKcJhtoeU,9443
+opik_optimizer/task_evaluator.py,sha256=7N254DU0UkWJ5saQ5AmYEsHHSrychAJtedmmjNsCOnI,5081
 opik_optimizer/data/context7_eval.jsonl,sha256=vPR3XRfI0UbZ1hgUGaOdpraFT99RDLU1YWuPFLLQz40,1757
 opik_optimizer/data/hotpot-500.json,sha256=YXxCtuvYvxSu5u0y4559a6b1qwgAYsWzT_SUKv_21ew,76862
 opik_optimizer/datasets/__init__.py,sha256=V4LVDOaRjwzaYvhdQ3V6CAwFaeKnxyTV1lp_ES9Z31E,691
@@ -27,29 +28,28 @@ opik_optimizer/demo/__init__.py,sha256=KSpFYhzN7fTmLEsIaciRHwxcJDeAiX5NDmYLdPsfp
 opik_optimizer/demo/cache.py,sha256=CwjdmVjokVxmPXvgfOutZK8e0sV-PIUz3ou6ODXZBts,3738
 opik_optimizer/demo/datasets.py,sha256=idod4NYHw1IbxhA8c0XVFD_pGpMZagNGNZuEYDTbbMM,2357
 opik_optimizer/evolutionary_optimizer/__init__.py,sha256=bDa6FZR9Y_a5z337I4EtvaB69jB542P4dbruhYPHCEU,95
-opik_optimizer/evolutionary_optimizer/crossover_ops.py,sha256=7kMvAWOiEA0R5PQMRdnLqbS1uCmIDVzLppNSsPsIO7o,7740
-opik_optimizer/evolutionary_optimizer/evaluation_ops.py,sha256=euVbFgrG37zj2GfcjAPMvPtz-52QljFR76OfChYmsKY,4795
-opik_optimizer/evolutionary_optimizer/evolutionary_optimizer.py,sha256=2gEny7Q1PMcwzaDOJ2GrXMKZAZThuhV9tGg92JhQ9lI,46260
+opik_optimizer/evolutionary_optimizer/crossover_ops.py,sha256=M-TsQv8EHKt_RiKoEPYTtiP_HL588AyyTuoXNsQpaVA,8883
+opik_optimizer/evolutionary_optimizer/evaluation_ops.py,sha256=XlPVJG_3R0GeYKOHTCdhBE4TvOBMRvyHlXwG2xvroD4,5511
+opik_optimizer/evolutionary_optimizer/evolutionary_optimizer.py,sha256=TVNHtQMZdGL4if_PK4f3230Rg0NFR87kOf0sibOeqbY,48162
 opik_optimizer/evolutionary_optimizer/helpers.py,sha256=yWYW5JyVbr2smDByc9yaHCYbUS6cw35RBI7lM3pT69A,607
-opik_optimizer/evolutionary_optimizer/llm_support.py,sha256=JeghAOwT_nYyOjdUi-xEiDvG-dW0C87UBzGz_xMdHl8,5438
-opik_optimizer/evolutionary_optimizer/mcp.py,sha256=QhRPsxbtZJKzhJouJOuNbrNGGjgY4JAMmo-UYiNa0WQ,7794
-opik_optimizer/evolutionary_optimizer/mutation_ops.py,sha256=Jj4q297z054LSI0udZmeH_jXQMAxml2_qBoxSIfXNBs,12643
-opik_optimizer/evolutionary_optimizer/population_ops.py,sha256=-33oN2aPTF_upJLYDVUTNm1c5bMzWy2krQ3alFCrJlM,10101
+opik_optimizer/evolutionary_optimizer/mcp.py,sha256=OHZ__q4vVInli8qowNh-1uD76dJ871wY6NkU4XhfxAA,8067
+opik_optimizer/evolutionary_optimizer/mutation_ops.py,sha256=ybhammJsY_SWIBsdZlom9H4Uy7-efbukma1j1-75oY4,13196
+opik_optimizer/evolutionary_optimizer/population_ops.py,sha256=ybNFUpfZgOeuWF5IGdtVRLFkiR4H9WsDrsGMPVj3Rk8,10992
 opik_optimizer/evolutionary_optimizer/prompts.py,sha256=am1nL8oqw3TOVVBDaDn5EoWkjxufEiMQ7E_54Uw8m3s,16204
-opik_optimizer/evolutionary_optimizer/reporting.py,sha256=xzvHK2m0Kdf7hhrrdRxXbp-qt8d8j69nnSBIOzhLSms,12090
+opik_optimizer/evolutionary_optimizer/reporting.py,sha256=AxhdiwUSEchKepu8eW6J5DiUYHXKvMDPqVQkrnrt1No,12048
 opik_optimizer/evolutionary_optimizer/style_ops.py,sha256=XmGFS5s2Qr2DJMZVVsI_C6LqJ5zoyxpeWAtGmdg3TnA,3082
 opik_optimizer/few_shot_bayesian_optimizer/__init__.py,sha256=VuH7FOROyGcjMPryejtZC-5Y0QHlVTFLTGUDgNqRAFw,113
-opik_optimizer/few_shot_bayesian_optimizer/few_shot_bayesian_optimizer.py,sha256=fiCp7_aPpoEf8afBhFo98z1pl1Z50Rs0b-lS6Rij2Ns,27897
-opik_optimizer/few_shot_bayesian_optimizer/reporting.py,sha256=OMpLG4xsM6K7oQcP_nbnky47NklVsowNDlK6WliZM10,6311
+opik_optimizer/few_shot_bayesian_optimizer/few_shot_bayesian_optimizer.py,sha256=3lFUcAiRCM6GUV2Aa9rbvF36nFQ1nFRK-ZC2_dkdGIU,26639
+opik_optimizer/few_shot_bayesian_optimizer/reporting.py,sha256=xk7gKaoTrlp1WDpW3mB5Irzty5Z5l9SJygO3PaamOvU,6283
 opik_optimizer/gepa_optimizer/__init__.py,sha256=XcPah5t4mop7UCFo69E9l45Mem49-itqkQT7_J1aWOA,71
 opik_optimizer/gepa_optimizer/adapter.py,sha256=KzPa4koq7aJhALMAOKPxAO4yWuEy_YbW7tGnqny3Hfo,5139
-opik_optimizer/gepa_optimizer/gepa_optimizer.py,sha256=HBjikhce3K4VIaiIXs7eSagmRyFPdY8h4seoW9F3nQE,26481
-opik_optimizer/gepa_optimizer/reporting.py,sha256=F0cxYSjRuFAszgi3rgqwH1A-KH26kZOLtENP7x1xrQs,5154
+opik_optimizer/gepa_optimizer/gepa_optimizer.py,sha256=rWD92KauJbnkXbsftzhTDkBjmgzlwGZcm6t7F-ceh_Q,27055
+opik_optimizer/gepa_optimizer/reporting.py,sha256=ZyD4bwiW6BxmPb-966u3iDAfBsiO56kO2VBxHtNmL-Q,11050
 opik_optimizer/hierarchical_reflective_optimizer/__init__.py,sha256=9qM3kvfAaFy-Y6Tg19MXHJxpnF5DJQQwzr6oNsxaRBM,133
-opik_optimizer/hierarchical_reflective_optimizer/hierarchical_reflective_optimizer.py,sha256=Fs83ztOuPS8mkFvJAVmYok15DaXTk4Jqpoa9ImRl2t4,27256
-opik_optimizer/hierarchical_reflective_optimizer/hierarchical_root_cause_analyzer.py,sha256=GSIXUBxoS9LFnCXopS6B6wLSpmCYXA8Cv6ytELgEBoc,12709
-opik_optimizer/hierarchical_reflective_optimizer/prompts.py,sha256=XcOEI9eeEbTgKFsFiRWxvHdaByQkiN02bH2gTl3HX-Y,3853
-opik_optimizer/hierarchical_reflective_optimizer/reporting.py,sha256=LpHv_WBZCg2a0RhZaGwUmCch_-Dfk_rpuMxTckJMWTU,23234
+opik_optimizer/hierarchical_reflective_optimizer/hierarchical_reflective_optimizer.py,sha256=fhB68XrGNgaHfPwV1JDow-MiAT-jhKDT_Kf-mLLzk0o,27775
+opik_optimizer/hierarchical_reflective_optimizer/hierarchical_root_cause_analyzer.py,sha256=0D5wgx04jZvTJ0Yjqm0jtQvkjrGBB73qgcsSwLBpnv0,13814
+opik_optimizer/hierarchical_reflective_optimizer/prompts.py,sha256=8TsLsJo7KPUNFkxSVGXTpVnr9ax4oTosImky0nlEI40,4376
+opik_optimizer/hierarchical_reflective_optimizer/reporting.py,sha256=frbqEOGsiK-TRPJTtcLhHjPJtQaj4T60cq97QEgcDJ0,29053
 opik_optimizer/hierarchical_reflective_optimizer/types.py,sha256=bS-JAheX2FpJ4XAxoZi5PfjloG8L-B1LGQA1iLXZhW4,1031
 opik_optimizer/mcp_utils/__init__.py,sha256=BsWQT8nAa6JV6zcOD__OvPMepUS2IpJD4J2rnAXhpuU,710
 opik_optimizer/mcp_utils/mcp.py,sha256=UylgpTJsybszS433_kuTAgKH-PPde-VHjHVelMardFs,18466
@@ -57,30 +57,30 @@ opik_optimizer/mcp_utils/mcp_second_pass.py,sha256=p2Knlxg1CKIZVMBbdskdRDqw1BRrn
 opik_optimizer/mcp_utils/mcp_simulator.py,sha256=bLL7iVAGMRc8Mb2j_XpSjlkr6TvQLI90hkS4ifnwLqs,3427
 opik_optimizer/mcp_utils/mcp_workflow.py,sha256=R3lqufN35p-OJlGxIxAIOMIAvRTBLGXINzfpoVIq2nw,17885
 opik_optimizer/meta_prompt_optimizer/__init__.py,sha256=syiN2_fMm5iZDQezZCHYe-ZiGOIPlBkLt49Sa1kuR70,97
-opik_optimizer/meta_prompt_optimizer/meta_prompt_optimizer.py,sha256=F3bxrttYAIkzvjduLvVDpFd9xoZkIqqTgqpSjgYDMIw,51435
-opik_optimizer/meta_prompt_optimizer/reporting.py,sha256=Py30NDYFNPzb8XrCXibQRtBC3vjjViQG74uG-O6lhXE,7783
-opik_optimizer/mipro_optimizer/__init__.py,sha256=7sMq9OSWyjITqK7sVtkO9fhG1w6KRE8bN7V52CKaGvo,94
-opik_optimizer/mipro_optimizer/_lm.py,sha256=jgp_bamkG9og8nxVKs6J2qPi6BmTvJD3qVeiorRhszU,17004
-opik_optimizer/mipro_optimizer/_mipro_optimizer_v2.py,sha256=bQBJG3wFeNsOF_Yhklx53d4M8kqzBXQ22cOLANXjGJw,39315
-opik_optimizer/mipro_optimizer/mipro_optimizer.py,sha256=ei3gON1r0PiCNhJVJA3zhGS4C2QPZOZMivhc2CPJ_R8,27660
-opik_optimizer/mipro_optimizer/utils.py,sha256=pP3mvai_GQmwUhcchVOiW1xPI3LatpXclE_5XvBYwTw,2493
+opik_optimizer/meta_prompt_optimizer/meta_prompt_optimizer.py,sha256=Q7wZgqynai6eezTVRAJ-FWPrYP6HlFpwQdiObbnwu7o,51562
+opik_optimizer/meta_prompt_optimizer/reporting.py,sha256=jW3WwMz7ZP7jjpvJLVyVXvlKbyTgMJVTz6mFYJhsmPQ,7755
+opik_optimizer/mipro_optimizer/__init__.py,sha256=Hr5HJT7LBBtbCTqBM0CSrIyYxq-eMfI2vujzEkCejV4,63
+opik_optimizer/mipro_optimizer/_lm.py,sha256=LjFan3--gCaeYxWEKdHswCcea_9jC1nLHK5uXULv-c4,17008
+opik_optimizer/mipro_optimizer/mipro_optimizer_v2.py,sha256=Dt2ETxyQXdkVKE74Zu0D2K90tTnTTIPjtp6uV60HCrc,39212
+opik_optimizer/mipro_optimizer/utils.py,sha256=hzlXYOFjCx1Vc485oby5uqI6Xoqc39x2aEpPT2acsf0,2539
 opik_optimizer/optimization_config/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-opik_optimizer/optimization_config/chat_prompt.py,sha256=d3jwM1UvUeRQOSsYHa5GD842VO3JWjVDmB3ROUGp57c,7089
+opik_optimizer/optimization_config/chat_prompt.py,sha256=LJGR2o6DMYdTEU2cbhX9dyH-hd9RQ8ngFfJHttBJJo4,6998
 opik_optimizer/optimization_config/configs.py,sha256=EGacRNnl6TeWuf8RNsxpP6Nh5JhogjK-JxKllK8dQr0,413
 opik_optimizer/optimization_config/mappers.py,sha256=4uBoPaIvCo4bqt_w-4rJyVe2LMAP_W7p6xxnDmGT-Sk,1724
 opik_optimizer/parameter_optimizer/__init__.py,sha256=Eg-LEFBJqnOFw7i2B_YH27CoIGDPb5y_q1ar-ZpjtYo,308
-opik_optimizer/parameter_optimizer/parameter_optimizer.py,sha256=eDd9tFQinz2lKsEJtikCBVzSWMK4saI9bhUY2NtDEg0,14955
+opik_optimizer/parameter_optimizer/parameter_optimizer.py,sha256=B18H2ZtcaPmrjh7f7_vHA8UewXml_9bw5PDA3nBzgyE,21206
 opik_optimizer/parameter_optimizer/parameter_search_space.py,sha256=rgTNK8HPbdDiVm4GVX2QESTmQPhPFj4UkxqZfAy9JAA,4659
 opik_optimizer/parameter_optimizer/parameter_spec.py,sha256=HzYT_dHBTfZtx403mY-Epv_IEqn4kYuYBZ6QUdkFRiY,8064
+opik_optimizer/parameter_optimizer/reporting.py,sha256=-kEe9sQFdkUhxayEamXLR8ukyTLJrGsTs8pbJWmimQ4,4665
 opik_optimizer/parameter_optimizer/search_space_types.py,sha256=UajTA2QKikEWazokDNO7j141gc2WxxYYiDRnFFjXi6M,512
 opik_optimizer/parameter_optimizer/sensitivity_analysis.py,sha256=8KEMVMHsmcoiK21Cq1-We6_Pw_6LX9qBX9Az4-tmj_w,2146
 opik_optimizer/utils/__init__.py,sha256=Ee0SnTPOcwRwp93M6Lh-X913lfSIwnvCiYYh5cpdRQE,486
 opik_optimizer/utils/colbert.py,sha256=qSrzKUUGw7P92mLy4Ofug5pBGeTsHBLMJXlXSJSfKuo,8147
-opik_optimizer/utils/core.py,sha256=5GT1vp6fW8ICO42LHMX14BjR-xEb6afAKjM7b1Evx5M,15298
+opik_optimizer/utils/core.py,sha256=sL9I9kG1Gdjj0b3rBgPpXp7NUaUisD3_ITSkE7w5QhU,16014
 opik_optimizer/utils/dataset_utils.py,sha256=dqRUGOekjeNWL0J15R8xFwLyKJDJynJXzVyQmt8rhHA,1464
-opik_optimizer/utils/prompt_segments.py,sha256=1zUITSccJ82Njac1rmANzim4WWM6rVac61mfluS7lFE,5931
-opik_optimizer-2.1.2.dist-info/licenses/LICENSE,sha256=V-0VHJOBdcA_teT8VymvsBUQ1-CZU6yJRmMEjec_8tA,11372
-opik_optimizer-2.1.2.dist-info/METADATA,sha256=V2WXOFaF2tc7-xAHcdndLn8pcUp1JHxHEVONa0K8j78,12829
-opik_optimizer-2.1.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-opik_optimizer-2.1.2.dist-info/top_level.txt,sha256=ondOlpq6_yFckqpxoAHSfzZS2N-JfgmA-QQhOJfz7m0,15
-opik_optimizer-2.1.2.dist-info/RECORD,,
+opik_optimizer/utils/prompt_segments.py,sha256=eiLYT1iiPxtB7ArriN13-LgI5tID-v7MrjniTAxK2Lo,5904
+opik_optimizer-2.2.0.dist-info/licenses/LICENSE,sha256=V-0VHJOBdcA_teT8VymvsBUQ1-CZU6yJRmMEjec_8tA,11372
+opik_optimizer-2.2.0.dist-info/METADATA,sha256=O8dgJi_lpEyd2kDMCj-AZMK4QthqAaZiR5Afo5WwCZ8,12807
+opik_optimizer-2.2.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+opik_optimizer-2.2.0.dist-info/top_level.txt,sha256=ondOlpq6_yFckqpxoAHSfzZS2N-JfgmA-QQhOJfz7m0,15
+opik_optimizer-2.2.0.dist-info/RECORD,,

opik_optimizer/evolutionary_optimizer/llm_support.py

@@ -1,136 +0,0 @@
-from typing import Any, TYPE_CHECKING
-
-import logging
-import os
-import time
-import random
-
-import litellm
-from litellm import exceptions as litellm_exceptions
-from litellm.caching import Cache
-from litellm.types.caching import LiteLLMCacheType
-from opik.evaluation.models.litellm import opik_monitor as opik_litellm_monitor
-
-from .. import _throttle
-
-
-logger = logging.getLogger(__name__)
-
-
-# Configure LiteLLM cache with safe fallback
-try:
-    # Prefer a disk cache in a user-writable location
-    cache_dir = os.path.join(os.path.expanduser("~"), ".cache", "litellm")
-    os.makedirs(cache_dir, exist_ok=True)
-    litellm.cache = Cache(type=LiteLLMCacheType.DISK, cache_dir=cache_dir)
-except (PermissionError, OSError, FileNotFoundError):
-    # Fall back to in-memory cache to avoid disk timeouts/locks
-    litellm.cache = Cache(type=LiteLLMCacheType.MEMORY)
-
-_rate_limiter = _throttle.get_rate_limiter_for_current_opik_installation()
-
-
-class LlmSupport:
-    if TYPE_CHECKING:
-        model: str
-        llm_call_counter: int
-        project_name: str | None
-        disable_litellm_monitoring: bool
-        temperature: float
-        max_tokens: int
-        top_p: float
-        frequency_penalty: float
-        presence_penalty: float
-
-        def increment_llm_counter(self) -> None: ...
-
-    @_throttle.rate_limited(_rate_limiter)
-    def _call_model(
-        self,
-        messages: list[dict[str, str]],
-        is_reasoning: bool = False,
-        optimization_id: str | None = None,
-    ) -> str:
-        """Call the model with the given prompt and return the response string."""
-        # Build base call params
-        llm_config_params: dict[str, Any] = {
-            "temperature": getattr(self, "temperature", 0.3),
-            "max_tokens": getattr(self, "max_tokens", 1000),
-            "top_p": getattr(self, "top_p", 1.0),
-            "frequency_penalty": getattr(self, "frequency_penalty", 0.0),
-            "presence_penalty": getattr(self, "presence_penalty", 0.0),
-        }
-
-        # Add Opik metadata unless disabled
-        try:
-            disable_monitoring_env = os.getenv(
-                "OPIK_OPTIMIZER_DISABLE_LITELLM_MONITORING", "0"
-            )
-            disable_monitoring = getattr(
-                self, "disable_litellm_monitoring", False
-            ) or disable_monitoring_env.lower() in ("1", "true", "yes")
-
-            if not disable_monitoring:
-                metadata_for_opik: dict[str, Any] = {}
-                pn = getattr(self, "project_name", None)
-                if pn:
-                    metadata_for_opik["project_name"] = pn
-                    metadata_for_opik["opik"] = {"project_name": pn}
-                if optimization_id and "opik" in metadata_for_opik:
-                    metadata_for_opik["opik"]["optimization_id"] = optimization_id
-                metadata_for_opik["optimizer_name"] = self.__class__.__name__
-                metadata_for_opik["opik_call_type"] = (
-                    "reasoning" if is_reasoning else "evaluation_llm_task_direct"
-                )
-                if metadata_for_opik:
-                    llm_config_params["metadata"] = metadata_for_opik
-
-                # Try to add Opik monitoring callbacks; fall back silently on failure
-                llm_config_params = (
-                    opik_litellm_monitor.try_add_opik_monitoring_to_params(  # type: ignore
-                        llm_config_params.copy()
-                    )
-                )
-        except Exception as e:
-            logger.debug(f"Skipping Opik-LiteLLM monitoring setup: {e}")
-
-        # Retry policy for transient errors
-        max_retries = int(os.getenv("OPIK_OPTIMIZER_LITELLM_MAX_RETRIES", "3"))
-        base_sleep = float(os.getenv("OPIK_OPTIMIZER_LITELLM_BACKOFF", "0.5"))
-
-        for attempt in range(max_retries + 1):
-            try:
-                logger.debug(
-                    f"Calling model '{self.model}' with messages: {messages}, params: {llm_config_params} (attempt {attempt + 1})"
-                )
-                response = litellm.completion(
-                    model=self.model, messages=messages, **llm_config_params
-                )
-                self.increment_llm_counter()
-                return response.choices[0].message.content
-            except (
-                litellm_exceptions.RateLimitError,
-                litellm_exceptions.APIConnectionError,
-                litellm_exceptions.InternalServerError,
-            ) as e:
-                if attempt < max_retries:
-                    sleep_s = min(10.0, base_sleep * (2**attempt)) + random.uniform(
-                        0, 0.25
-                    )
-                    logger.warning(
-                        f"LiteLLM transient error ({type(e).__name__}): {e}. Retrying in {sleep_s:.2f}s..."
-                    )
-                    time.sleep(sleep_s)
-                    continue
-                logger.error(f"LiteLLM error (final attempt): {e}")
-                raise
-            except litellm_exceptions.ContextWindowExceededError as e:
-                logger.error(f"LiteLLM Context Window Exceeded Error: {e}")
-                raise
-            except Exception as e:
-                logger.error(
-                    f"Error calling model '{self.model}': {type(e).__name__} - {e}"
-                )
-                raise
-        # Should never reach here
-        raise RuntimeError("LLM call did not return a response and did not raise")