verifiers 0.4.1.dev58__tar.gz → 0.4.1.dev59__tar.gz

{verifiers-0.4.1.dev58 → verifiers-0.4.1.dev59}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.4.1.dev58
+Version: 0.4.1.dev59
 Summary: Verifiers: Environments for LLM Reinforcement Learning
 Project-URL: Homepage, https://github.com/primeintellect-ai/verifiers
 Project-URL: Documentation, https://github.com/primeintellect-ai/verifiers

{verifiers-0.4.1.dev58 → verifiers-0.4.1.dev59}/verifiers/v1/task.py

@@ -14,6 +14,7 @@ import hashlib
 import inspect
 import json
 import logging
+import math
 from collections.abc import Callable, Mapping
 from typing import TYPE_CHECKING, Any, ClassVar, Generic, Self
 
@@ -24,7 +25,7 @@ from verifiers.v1.configs.harness import SkillSource
 from verifiers.v1.configs.runtime import NetworkPolicyConfig
 from verifiers.v1.configs.scoring import ScoringConfig
 from verifiers.v1.configs.task import TaskConfig
-from verifiers.v1.errors import TaskError, boundary
+from verifiers.v1.errors import TaskError, UnscoredError, boundary
 from verifiers.v1.state import StateT
 from verifiers.v1.types import Messages, content_text
 from verifiers.v1.utils.artifacts import MAX_ARTIFACT_BYTES, Artifact
@@ -232,6 +233,15 @@ class Task(Generic[DataT, StateT, ConfigT]):
         trace.info.pop("judge_calls", None)
         trace.rewards, trace.metrics, trace.extra_usage = {}, {}, []
         await self.score(trace)
+        unscored = [
+            name
+            for name, reward in trace.rewards.items()
+            if reward is not None and math.isnan(reward.score)
+        ]
+        if unscored:
+            for name in unscored:
+                trace.rewards[name] = None
+            raise UnscoredError(f"reward {', '.join(unscored)} scored nothing (nan)")
 
     def defer_scoring(self) -> Self:
         """An independent copy whose task signals are deferred.