aiqtoolkit 1.2.0a20250621__py3-none-any.whl → 1.2.0a20250623__py3-none-any.whl

aiq/eval/evaluate.py

@@ -32,6 +32,7 @@ from aiq.eval.evaluator.evaluator_model import EvalInput
 from aiq.eval.evaluator.evaluator_model import EvalInputItem
 from aiq.eval.evaluator.evaluator_model import EvalOutput
 from aiq.eval.utils.output_uploader import OutputUploader
+from aiq.eval.utils.weave_eval import WeaveEvaluationIntegration
 from aiq.runtime.session import AIQSessionManager
 
 logger = logging.getLogger(__name__)
@@ -54,7 +55,7 @@ class EvaluationRun:  # pylint: disable=too-many-public-methods
 
         # Helpers
         self.intermediate_step_adapter: IntermediateStepAdapter = IntermediateStepAdapter()
-
+        self.weave_eval: WeaveEvaluationIntegration = WeaveEvaluationIntegration()
         # Metadata
         self.eval_input: EvalInput | None = None
         self.workflow_interrupted: bool = False
@@ -138,6 +139,8 @@ class EvaluationRun:  # pylint: disable=too-many-public-methods
                 item.output_obj = output
                 item.trajectory = self.intermediate_step_adapter.validate_intermediate_steps(intermediate_steps)
 
+                self.weave_eval.log_prediction(item, output)
+
         async def wrapped_run(item: EvalInputItem) -> None:
             await run_one(item)
             pbar.update(1)
@@ -268,11 +271,15 @@ class EvaluationRun:  # pylint: disable=too-many-public-methods
                    "`eval` with the --skip_completed_entries flag.")
             logger.warning(msg)
 
+        self.weave_eval.log_summary(self.evaluation_results)
+
     async def run_single_evaluator(self, evaluator_name: str, evaluator: Any):
         """Run a single evaluator and store its results."""
         try:
             eval_output = await evaluator.evaluate_fn(self.eval_input)
             self.evaluation_results.append((evaluator_name, eval_output))
+
+            await self.weave_eval.alog_score(eval_output, evaluator_name)
         except Exception as e:
             logger.exception("An error occurred while running evaluator %s: %s", evaluator_name, e, exc_info=True)
 
@@ -289,6 +296,9 @@ class EvaluationRun:  # pylint: disable=too-many-public-methods
         except Exception as e:
             logger.exception("An error occurred while running evaluators: %s", e, exc_info=True)
             raise
+        finally:
+            # Finish prediction loggers in Weave
+            await self.weave_eval.afinish_loggers()
 
     def apply_overrides(self):
         from aiq.cli.cli_utils.config_override import load_and_override_config
@@ -362,6 +372,11 @@ class EvaluationRun:  # pylint: disable=too-many-public-methods
 
         # Run workflow and evaluate
         async with WorkflowEvalBuilder.from_config(config=config) as eval_workflow:
+            # Initialize Weave integration
+            self.weave_eval.initialize_client()
+            if self.weave_eval.client:
+                self.weave_eval.initialize_logger(self.eval_input, config)
+
             if self.config.endpoint:
                 await self.run_workflow_remote()
             else:

aiq/eval/utils/weave_eval.py

@@ -0,0 +1,135 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import logging
+from typing import Any
+from typing import List
+
+from aiq.eval.evaluator.evaluator_model import EvalInput
+from aiq.eval.evaluator.evaluator_model import EvalInputItem
+from aiq.eval.evaluator.evaluator_model import EvalOutput
+
+logger = logging.getLogger(__name__)
+
+
+class WeaveEvaluationIntegration:  # pylint: disable=too-many-public-methods
+    """
+    Class to handle all Weave integration functionality.
+    """
+
+    def __init__(self):
+        self.available = False
+        self.client = None
+        self.eval_logger = None
+        self.pred_loggers = {}
+
+        try:
+            from weave.flow.eval_imperative import EvaluationLogger
+            from weave.flow.eval_imperative import ScoreLogger
+            from weave.trace.context import weave_client_context
+            self.EvaluationLogger = EvaluationLogger
+            self.ScoreLogger = ScoreLogger
+            self.weave_client_context = weave_client_context
+            self.available = True
+        except ImportError:
+            self.available = False
+            # we simply don't do anything if weave is not available
+            pass
+
+    def initialize_client(self):
+        """Initialize the Weave client if available."""
+        if not self.available:
+            return False
+
+        try:
+            self.client = self.weave_client_context.require_weave_client()
+            return self.client is not None
+        except Exception:
+            self.client = None
+            return False
+
+    def initialize_logger(self, eval_input: EvalInput, config: Any):
+        """Initialize the Weave evaluation logger."""
+        if not self.client:
+            return False
+
+        try:
+            weave_dataset = [
+                item.model_dump(exclude={"output_obj", "trajectory"}) for item in eval_input.eval_input_items
+            ]
+            config_dict = config.model_dump(mode="json")
+            # TODO: make this configurable
+            config_dict["name"] = "aiqtoolkit-eval"
+            self.eval_logger = self.EvaluationLogger(model=config_dict, dataset=weave_dataset)
+            self.pred_loggers = {}
+
+            del weave_dataset
+            del config_dict
+            return True
+        except Exception as e:
+            self.eval_logger = None
+            logger.warning("Failed to initialize Weave `EvaluationLogger`: %s", e)
+
+            return False
+
+    def log_prediction(self, item: EvalInputItem, output: Any):
+        """Log a prediction to Weave."""
+        if not self.eval_logger:
+            return
+
+        pred_logger = self.eval_logger.log_prediction(inputs=item.model_dump(exclude={"output_obj", "trajectory"}),
+                                                      output=output)
+        self.pred_loggers[item.id] = pred_logger
+
+    async def alog_score(self, eval_output: EvalOutput, evaluator_name: str):
+        """Log scores for evaluation outputs."""
+        if not self.eval_logger:
+            return
+
+        for eval_output_item in eval_output.eval_output_items:
+            if eval_output_item.id in self.pred_loggers:
+                await self.pred_loggers[eval_output_item.id].alog_score(
+                    scorer=evaluator_name,
+                    score=eval_output_item.score,
+                )
+
+    async def afinish_loggers(self):
+        """Finish all prediction loggers."""
+        if not self.eval_logger:
+            return
+
+        async def _finish_one(pred_logger):
+            if hasattr(pred_logger, '_has_finished') and not pred_logger._has_finished:
+                return
+            # run the *blocking* finish() in a thread so we don’t nest loops
+            await asyncio.to_thread(pred_logger.finish)
+
+        await asyncio.gather(*[_finish_one(pl) for pl in self.pred_loggers.values()])
+
+    def log_summary(self, evaluation_results: List[tuple[str, EvalOutput]]):
+        """Log summary statistics to Weave."""
+        if not self.eval_logger:
+            return
+
+        summary = {}
+        for evaluator_name, eval_output in evaluation_results:
+            # Calculate average score for this evaluator
+            scores = [item.score for item in eval_output.eval_output_items if item.score is not None]
+            if scores:
+                summary[f"{evaluator_name}_avg"] = sum(scores) / len(scores)
+
+        # Log the summary to finish the evaluation
+        self.eval_logger.log_summary(summary)

{aiqtoolkit-1.2.0a20250621.dist-info → aiqtoolkit-1.2.0a20250623.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: aiqtoolkit
-Version: 1.2.0a20250621
+Version: 1.2.0a20250623
 Summary: NVIDIA Agent Intelligence toolkit
 Author: NVIDIA Corporation
 Maintainer: NVIDIA Corporation

{aiqtoolkit-1.2.0a20250621.dist-info → aiqtoolkit-1.2.0a20250623.dist-info}/RECORD

@@ -107,7 +107,7 @@ aiq/embedder/openai_embedder.py,sha256=5FO3xsyNvEmbLBsZb3xsCpbN1Soxio4yf4b5gTPVx
 aiq/embedder/register.py,sha256=3MTZrfNQKp6AZTbfaA-PpTnyXiMyu-8HH9JnDCC0v9o,978
 aiq/eval/__init__.py,sha256=Xs1JQ16L9btwreh4pdGKwskffAw1YFO48jKrU4ib_7c,685
 aiq/eval/config.py,sha256=IlOr2o618kbkXP0G1F-AklZfsKYVos9UB4Dvlxf66xk,1431
-aiq/eval/evaluate.py,sha256=AGEvmagd43jLq0aE_yNs_FFPFxVJEx49cu6Fl3WeQqA,17270
+aiq/eval/evaluate.py,sha256=VdVdB_CV842gIV4diHciJ1qrof5_N3H8I16WwracCsQ,17940
 aiq/eval/intermediate_step_adapter.py,sha256=4cSsGgFBvNjXnclk5FvZnQaFEdeulp7VEdRWKLcREAQ,4498
 aiq/eval/register.py,sha256=QOHJqA2CQixeWMC9InyKbzXo1jByvrntD_m9-2Mvg9k,1076
 aiq/eval/remote_workflow.py,sha256=Fb7Z6gdP2L_gqyWB--AEWfcXe9xPpQ_hPsf9lmqGXjI,5524
@@ -134,6 +134,7 @@ aiq/eval/tunable_rag_evaluator/register.py,sha256=q4p2rFyMzWmaINJc961ZV4jzIlAN4G
 aiq/eval/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 aiq/eval/utils/output_uploader.py,sha256=SaQbZPkw-Q0H7t5yG60Kh-p1cflR7gPklVkilC4uPbU,5141
 aiq/eval/utils/tqdm_position_registry.py,sha256=9CtpCk1wtYCSyieHPaSp8nlZu6EcNUOaUz2RTqfekrA,1286
+aiq/eval/utils/weave_eval.py,sha256=yIdlp4UdCPgwFYJNJon5eZD1d99E-6dcmfVg6B-4RKE,5076
 aiq/front_ends/__init__.py,sha256=Xs1JQ16L9btwreh4pdGKwskffAw1YFO48jKrU4ib_7c,685
 aiq/front_ends/register.py,sha256=OKv1xi-g8WHtUMuIPhwjG6wOYqaGDD-Q9vDtKtT9d1Y,889
 aiq/front_ends/console/__init__.py,sha256=Xs1JQ16L9btwreh4pdGKwskffAw1YFO48jKrU4ib_7c,685
@@ -309,10 +310,10 @@ aiq/utils/reactive/base/observer_base.py,sha256=UAlyAY_ky4q2t0P81RVFo2Bs_R7z5Nde
 aiq/utils/reactive/base/subject_base.py,sha256=Ed-AC6P7cT3qkW1EXjzbd5M9WpVoeN_9KCe3OM3FLU4,2521
 aiq/utils/settings/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 aiq/utils/settings/global_settings.py,sha256=U9TCLdoZsKq5qOVGjREipGVv9e-FlStzqy5zv82_VYk,7454
-aiqtoolkit-1.2.0a20250621.dist-info/licenses/LICENSE-3rd-party.txt,sha256=8o7aySJa9CBvFshPcsRdJbczzdNyDGJ8b0J67WRUQ2k,183936
-aiqtoolkit-1.2.0a20250621.dist-info/licenses/LICENSE.md,sha256=QwcOLU5TJoTeUhuIXzhdCEEDDvorGiC6-3YTOl4TecE,11356
-aiqtoolkit-1.2.0a20250621.dist-info/METADATA,sha256=D14wpPMZp_1mnHcSmKNtKAky2uL_D_5b2i0rs39bkiM,20274
-aiqtoolkit-1.2.0a20250621.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-aiqtoolkit-1.2.0a20250621.dist-info/entry_points.txt,sha256=gRlPfR5g21t328WNEQ4CcEz80S1sJNS8A7rMDYnzl4A,452
-aiqtoolkit-1.2.0a20250621.dist-info/top_level.txt,sha256=fo7AzYcNhZ_tRWrhGumtxwnxMew4xrT1iwouDy_f0Kc,4
-aiqtoolkit-1.2.0a20250621.dist-info/RECORD,,
+aiqtoolkit-1.2.0a20250623.dist-info/licenses/LICENSE-3rd-party.txt,sha256=8o7aySJa9CBvFshPcsRdJbczzdNyDGJ8b0J67WRUQ2k,183936
+aiqtoolkit-1.2.0a20250623.dist-info/licenses/LICENSE.md,sha256=QwcOLU5TJoTeUhuIXzhdCEEDDvorGiC6-3YTOl4TecE,11356
+aiqtoolkit-1.2.0a20250623.dist-info/METADATA,sha256=M98GDq-TQ5hxx-6C2mgkvNJD-NPsksihuT96qOukjEE,20274
+aiqtoolkit-1.2.0a20250623.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+aiqtoolkit-1.2.0a20250623.dist-info/entry_points.txt,sha256=gRlPfR5g21t328WNEQ4CcEz80S1sJNS8A7rMDYnzl4A,452
+aiqtoolkit-1.2.0a20250623.dist-info/top_level.txt,sha256=fo7AzYcNhZ_tRWrhGumtxwnxMew4xrT1iwouDy_f0Kc,4
+aiqtoolkit-1.2.0a20250623.dist-info/RECORD,,