verifiers 0.3.2.dev144__tar.gz → 0.3.2.dev145__tar.gz

{verifiers-0.3.2.dev144 → verifiers-0.3.2.dev145}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev144
+Version: 0.3.2.dev145
 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-0.3.2.dev144 → verifiers-0.3.2.dev145}/verifiers/v1/cli/replay.py

@@ -148,7 +148,7 @@ async def run_replay(config: ReplayConfig, source: Path, out: Path) -> list[Trac
         async with sem or contextlib.nullcontext():
             st.start = time.time()
             # Generation failures have no complete transcript to score.
-            if trace.stop_condition == "error":
+            if trace.has_error:
                 st.state, st.detail, st.end = "skipped", "rollout errored", time.time()
             else:
                 st.state = "running"

{verifiers-0.3.2.dev144 → verifiers-0.3.2.dev145}/verifiers/v1/errors.py

@@ -90,6 +90,14 @@ async def boundary(error_cls: type[RolloutError], what: str) -> AsyncIterator[No
         raise error_cls(f"{what}: {type(e).__name__}: {e}") from e
 
 
+def stop_condition(error: BaseException) -> str:
+    """The stop condition a recorded error leaves on the trace: `<boundary>_error` for
+    a typed rollout error (`SandboxError` -> `sandbox_error`), `error` for any other."""
+    if isinstance(error, RolloutError):
+        return f"{type(error).__name__.removesuffix('Error').lower()}_error"
+    return "error"
+
+
 def _provider_status(e: OpenAIError | str) -> int:
     """The HTTP status to surface for an SDK error: the provider's own for an HTTP status error, a
     retryable 5xx for a transport/timeout fault, else 502."""

{verifiers-0.3.2.dev144 → verifiers-0.3.2.dev145}/verifiers/v1/trace.py

@@ -17,7 +17,7 @@ if TYPE_CHECKING:
 
 from verifiers.v1 import graph
 from verifiers.v1.configs.agent import AgentConfig, WireAgentConfig
-from verifiers.v1.errors import ProviderError
+from verifiers.v1.errors import ProviderError, stop_condition
 from verifiers.v1.graph import RECORD_FLOAT_DECIMALS, MessageNode
 from verifiers.v1.runtimes import RuntimeInfo
 from verifiers.v1.semantic import ACPInfo, ParentLink, SemanticEdgeSet
@@ -781,7 +781,8 @@ class Trace(BaseModel, Generic[DataT, StateT, AgentConfigT]):
         self.stop(f"{stage}_timeout", override=True)
 
     def record_error(self, error: Exception) -> None:
-        """Record an error, and stop the trace as failed."""
+        """Record an error, and stop the trace as failed: `<boundary>_error` for a
+        typed rollout error (`errors.stop_condition`), `error` for any other."""
         self.errors.append(
             Error(
                 type=type(error).__name__,
@@ -795,7 +796,7 @@ class Trace(BaseModel, Generic[DataT, StateT, AgentConfigT]):
             )
         )
         self.ok = False
-        self.stop("error")
+        self.stop(stop_condition(error))
 
     def to_record(
         self, float_decimals: int | None = RECORD_FLOAT_DECIMALS