themis-eval 0.2.2__py3-none-any.whl → 0.2.3__py3-none-any.whl

themis/_version.py

@@ -9,7 +9,7 @@ def _detect_version() -> str:
     try:
         return metadata.version("themis-eval")
     except metadata.PackageNotFoundError:  # pragma: no cover - local dev only
-        return "0.2.2"  # Fallback for development
+        return "0.2.3"  # Fallback for development
 
 
 __version__ = _detect_version()

themis/evaluation/pipelines/standard_pipeline.py

@@ -233,7 +233,8 @@ class EvaluationPipeline:
                             if reference is not None
                             else []
                         )
-                        metadata = {"sample_id": sample_id}
+                        # Preserve all task metadata for metrics, add sample_id
+                        metadata = {**record.task.metadata, "sample_id": sample_id}
                         extract_start = time.perf_counter()
                         item_scores_for_item: list[core_entities.MetricScore] = []
                         for metric in self._metrics:

themis/evaluation/strategies/attempt_aware_evaluation_strategy.py

@@ -37,13 +37,16 @@ class AttemptAwareEvaluationStrategy:
             grouped.setdefault(score.metric_name, []).append(score)
         for metric_name, group in grouped.items():
             value = sum(item.value for item in group) / len(group)
+            # Preserve original metadata from first score
+            base_metadata = group[0].metadata.copy() if group[0].metadata else {}
             aggregated.append(
                 core_entities.MetricScore(
                     metric_name=metric_name,
                     value=value,
                     metadata={
-                        "attempts": len(group),
-                        "sample_id": group[0].metadata.get("sample_id"),
+                        **base_metadata,  # Preserve all original metadata
+                        "attempts": len(group),  # Add aggregation-specific field
+                        "sample_id": base_metadata.get("sample_id"),
                     },
                     details={},
                 )

themis/evaluation/strategies/judge_evaluation_strategy.py

@@ -49,6 +49,8 @@ class JudgeEvaluationStrategy:
                 counts = Counter(labels)
                 agreement = max(counts.values()) / max(1, len(labels))
 
+            # Preserve original metadata from first score
+            base_metadata = group[0].metadata.copy() if group[0].metadata else {}
             aggregated.append(
                 core_entities.MetricScore(
                     metric_name=metric_name,
@@ -58,7 +60,10 @@ class JudgeEvaluationStrategy:
                         "agreement": agreement,
                         "labels": labels,
                     },
-                    metadata={"sample_id": group[0].metadata.get("sample_id")},
+                    metadata={
+                        **base_metadata,  # Preserve all original metadata
+                        "sample_id": base_metadata.get("sample_id"),
+                    },
                 )
             )
         return aggregated

themis/generation/plan.py

@@ -88,9 +88,20 @@ class GenerationPlan:
         }
         if dataset_id is not None:
             metadata["dataset_id"] = dataset_id
-        for field_name in self.metadata_fields:
-            if field_name in row:
-                metadata[field_name] = row[field_name]
+        
+        # If metadata_fields is explicitly specified, use it as a filter (existing behavior)
+        # Otherwise, include all fields by default (new behavior for custom metrics)
+        if self.metadata_fields:
+            # Explicit filter - only include specified fields
+            for field_name in self.metadata_fields:
+                if field_name in row:
+                    metadata[field_name] = row[field_name]
+        else:
+            # No filter - include all fields except those used for other purposes
+            for field_name, field_value in row.items():
+                if field_name not in (self.dataset_id_field, self.reference_field):
+                    metadata[field_name] = field_value
+        
         return metadata
 
     def _build_reference(
@@ -209,9 +220,20 @@ class CartesianExpansionStrategy:
         }
         if dataset_id is not None:
             metadata["dataset_id"] = dataset_id
-        for field_name in context.metadata_fields:
-            if field_name in row:
-                metadata[field_name] = row[field_name]
+        
+        # If metadata_fields is explicitly specified, use it as a filter (existing behavior)
+        # Otherwise, include all fields by default (new behavior for custom metrics)
+        if context.metadata_fields:
+            # Explicit filter - only include specified fields
+            for field_name in context.metadata_fields:
+                if field_name in row:
+                    metadata[field_name] = row[field_name]
+        else:
+            # No filter - include all fields except those used for other purposes
+            for field_name, field_value in row.items():
+                if field_name not in (context.dataset_id_field, context.reference_field):
+                    metadata[field_name] = field_value
+        
         return metadata
 
     def _build_reference(

{themis_eval-0.2.2.dist-info → themis_eval-0.2.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: themis-eval
-Version: 0.2.2
+Version: 0.2.3
 Summary: Lightweight evaluation platform for LLM experiments
 Author: Pittawat Taveekitworachai
 License: MIT

{themis_eval-0.2.2.dist-info → themis_eval-0.2.3.dist-info}/RECORD

@@ -1,5 +1,5 @@
 themis/__init__.py,sha256=rQL3njf3i5lnAcmu0HuRzGGMELbA9xX21hzw4HrbIxw,1394
-themis/_version.py,sha256=y0Oqv0Je2udPmKCy5_D8Lib7GNLGxtLVp8b5WdavITg,378
+themis/_version.py,sha256=Tk4OCTQHYoZ61gm9JnkdgajR0vkBHbVm5OUjInzyJug,378
 themis/api.py,sha256=flZTbU-jRcbv7oXcfRKG4hkZjASmWlT52A4PghKj9G0,17700
 themis/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 themis/backends/__init__.py,sha256=RWM5SnV5FrS_cVjpHHeZZM_b9CgqBu1rPS5DlT5YQTY,578
@@ -88,7 +88,7 @@ themis/evaluation/metrics/nlp/meteor.py,sha256=QZT09s4aiUcVvDJDVPZYjzi5SxXdS2gn2
 themis/evaluation/metrics/nlp/rouge.py,sha256=YL05qluF-KsesHYFRfm5zELJlcvo6RvaKp7xKy6BuLI,4365
 themis/evaluation/pipelines/__init__.py,sha256=5YI1xaUULHisctFxrumN4XRpWYneoonX7nd9zBtsjvQ,384
 themis/evaluation/pipelines/composable_pipeline.py,sha256=nNP9MSvQQJvaSBw5_gO3FeyhGm9So2ZlGqh5qSvE8Ac,10905
-themis/evaluation/pipelines/standard_pipeline.py,sha256=nDd_bkqAVQxgwG9RK6G_fsgqwZth3058uG3p4QM0Dck,14650
+themis/evaluation/pipelines/standard_pipeline.py,sha256=GI5_ImebBuM6D8GpGKLoNq4p3JhTq-ocOThlah8RxME,14754
 themis/evaluation/statistics/__init__.py,sha256=TTrScTLAW7EHNq0nbjuJs6iP3_HgDx1yy3EtYXx5JCk,1257
 themis/evaluation/statistics/bootstrap.py,sha256=JUQ8rtzFvW2e41I2pLJ7pqgSEjuJ1r6McyYLI42At9g,2409
 themis/evaluation/statistics/confidence_intervals.py,sha256=CN5EO2gWiSITQubuWuPryngnGXhGwczY9kO3mcG6JVc,3676
@@ -97,10 +97,10 @@ themis/evaluation/statistics/effect_sizes.py,sha256=EWFVDilczpR8rR3_YurWy7QcjYcN
 themis/evaluation/statistics/hypothesis_tests.py,sha256=MVlVsY8wXifbBG5aSwauFShsQtIKqYREJApbriojS2o,10042
 themis/evaluation/statistics/types.py,sha256=hW0RYWs-G4C_njNl0ZGG9lJROgU2CfLWfnTQDWYmWuw,3685
 themis/evaluation/strategies/__init__.py,sha256=3f5LQkzlu3pRbN7dgDbdYOUNZTRexcn6f8D8I5-C724,439
-themis/evaluation/strategies/attempt_aware_evaluation_strategy.py,sha256=O3dlsQ2F0Ucv2Dhjz2Qf-jpPhwaVs3zrdQDRRu9du5w,1714
+themis/evaluation/strategies/attempt_aware_evaluation_strategy.py,sha256=MFcBdtK8rBeDXPFD2YWPSprez2iwSB-8yfyWhAlylug,1959
 themis/evaluation/strategies/default_evaluation_strategy.py,sha256=LShW-3Nxg_W4Ln-4qUvHJZqe5YMt64gHoK3uNJYLQNo,693
 themis/evaluation/strategies/evaluation_strategy.py,sha256=YFF-bXkz4Z52GuCw52FcklfEnf8dK8_z_I40DJRcmwE,669
-themis/evaluation/strategies/judge_evaluation_strategy.py,sha256=58pDB30y1VpM_1KPB6sGS0JImGZk5WTgnK9CKDF8N5k,2304
+themis/evaluation/strategies/judge_evaluation_strategy.py,sha256=IRSgnnD2R6qrjiOTyA_PIOHUfQj4FqutkU3pKMth0CQ,2562
 themis/experiment/__init__.py,sha256=dGranqpESugmmfbQlTU9efwspazW6j3vcmAKEtAoWZk,182
 themis/experiment/builder.py,sha256=AEjCDeSOI2B0i0PBjkfY1GUDNrYGTGiqPvt0SxnDQFo,5618
 themis/experiment/cache_manager.py,sha256=Fd8Qxifrmyn8f2zjAyPrLv-ZU4Dcp-MKo8-09BoW7tY,4361
@@ -121,7 +121,7 @@ themis/generation/agentic_runner.py,sha256=armBQBk7qZDBEwT8HqjIWomYDQm57NfrP5CZJ
 themis/generation/batching.py,sha256=ddpgpn1pq_EwipvTg-K4WcoSs3c2rbW37jEA5Pa_spo,7557
 themis/generation/clients.py,sha256=6apXCp_VNQosnpnmohTHOhHGXw-VZgsUyLds8MwtYUE,4910
 themis/generation/conversation_runner.py,sha256=kSZHwEvfqzxZ-eQYxmg5OkNZcgEHggZExjad6nBOeTM,7980
-themis/generation/plan.py,sha256=RmPIdefXkQMHYv5EWiilpx91I9a-svw31imvG0wV3fE,15961
+themis/generation/plan.py,sha256=k6_gdKFM12nrKz7ac1c5vTZsFanIKJJgyQ8IhvakDNQ,17158
 themis/generation/router.py,sha256=jZc0KFL483f8TrYtt9yxzFKs-T9CG2CoE2kfOQdHMEc,1082
 themis/generation/runner.py,sha256=pH4Dw77qskMQk3yxEkaHYAl1PItTofI7OXdvevnFiCA,8984
 themis/generation/strategies.py,sha256=hjqaVkNycFxJWh_edJ7ilBl7HS6bL-8pYm24zTfoAvg,2975
@@ -150,8 +150,8 @@ themis/utils/dashboard.py,sha256=2yiIu9_oENglTde_J3G1d5cpQ5VtSnfbUvdliw5Og1E,130
 themis/utils/logging_utils.py,sha256=buC64X-xOu-2SZ0wVkz3nCXzYVGiqKbxK-8DGSGsAdM,1173
 themis/utils/progress.py,sha256=HS0-yVbRT7Ai9zRlsJcex_OKP6dUiKx1vOp_IsobiHM,2097
 themis/utils/tracing.py,sha256=VTeiRjcW_B5fOOoSeAp37nrmlwP1DiqPcoe6OtIQ7dk,8468
-themis_eval-0.2.2.dist-info/licenses/LICENSE,sha256=K5FLE7iqn5-_6k1sf3IGy7w-Wx_Vdx3t0sOVJByNlF0,1076
-themis_eval-0.2.2.dist-info/METADATA,sha256=eOlF2Obimv_822azCt0vwhLaBz3CKsuvJPgDHMA3WFU,15235
-themis_eval-0.2.2.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
-themis_eval-0.2.2.dist-info/top_level.txt,sha256=QGIl4v-KB32upFS5UTXMJxHVX3vF7yBso82wJFI1Vbs,7
-themis_eval-0.2.2.dist-info/RECORD,,
+themis_eval-0.2.3.dist-info/licenses/LICENSE,sha256=K5FLE7iqn5-_6k1sf3IGy7w-Wx_Vdx3t0sOVJByNlF0,1076
+themis_eval-0.2.3.dist-info/METADATA,sha256=4N7tBOyUi8PAlFT2qJseKIABjHOzkFmLtfqVVUSFz84,15235
+themis_eval-0.2.3.dist-info/WHEEL,sha256=qELbo2s1Yzl39ZmrAibXA2jjPLUYfnVhUNTlyF1rq0Y,92
+themis_eval-0.2.3.dist-info/top_level.txt,sha256=QGIl4v-KB32upFS5UTXMJxHVX3vF7yBso82wJFI1Vbs,7
+themis_eval-0.2.3.dist-info/RECORD,,