verifiers 0.3.2.dev162__py3-none-any.whl → 0.3.2.dev163__py3-none-any.whl

verifiers/v1/tasksets/harbor/taskset.py

@@ -372,7 +372,8 @@ class HarborTask(Task[HarborData, State, HarborTaskConfig]):
     ) -> float | dict[str, float]:
         # By absolute path, in the runtime's configured workdir: Harbor execs the
         # script the same way, and scripts do grade the agent's work at `$PWD`.
-        await runtime.run(
+        # The reward file is authoritative even when the script exits nonzero.
+        result = await runtime.run(
             ["bash", "/tests/test.sh"], resolve_env(self.data.verifier_env)
         )
         scores = await self.read_reward_json(runtime)
@@ -393,9 +394,13 @@ class HarborTask(Task[HarborData, State, HarborTaskConfig]):
                 .decode()
                 .strip()
             )
-            return float(reward or 0)
-        except (SandboxError, OSError, ValueError):
-            return 0.0
+            return REWARD_JSON_ADAPTER.validate_python(float(reward))
+        except (SandboxError, OSError, ValueError) as exc:
+            raise TaskError(
+                "Harbor verifier produced no valid reward.json or reward.txt "
+                f"(exit {result.exit_code}): "
+                f"{(result.stderr or result.stdout).strip()[-500:]}"
+            ) from exc
 
     async def read_reward_json(
         self, runtime: Runtime

{verifiers-0.3.2.dev162.dist-info → verifiers-0.3.2.dev163.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev162
+Version: 0.3.2.dev163
 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.dev162.dist-info → verifiers-0.3.2.dev163.dist-info}/RECORD

@@ -161,7 +161,7 @@ verifiers/v1/tasksets/__init__.py,sha256=2ijE3qIlLoUA02g6iysOwWQylyBBQL1GhSpjv8_
 verifiers/v1/tasksets/harbor/__init__.py,sha256=JxThzHFYEkuSfQbVf3pEnjordgdfe4MU-QOsSSbyovU,326
 verifiers/v1/tasksets/harbor/compose.py,sha256=Ab-vpMPFXy7Y7C_82-IRznrlaN3FLEuFDz8w3LDsE-I,14094
 verifiers/v1/tasksets/harbor/env.py,sha256=-U-CeyeuWysXCNHi3cew0XtakUa7TqJVnVzElKtG41w,6356
-verifiers/v1/tasksets/harbor/taskset.py,sha256=HQXrCu2Gpx_UH6AwOtitSXw-OvtdFrDUDewlNUqXWEk,36070
+verifiers/v1/tasksets/harbor/taskset.py,sha256=IaGeCoH012wcnPmCauk4kgGRdXsDt1qpMnOBKuUkV3Q,36421
 verifiers/v1/tasksets/harbor/toolset.py,sha256=_7HXu-uuvqzhmJ1EX6c3KWLg0g0eRrLjCq1kBsZhIwo,3159
 verifiers/v1/tasksets/nemo_gym/__init__.py,sha256=654gAvG3Wd_Z_nh6mypeKn4s_IIQ0Ge6vcZvxTHC2_E,306
 verifiers/v1/tasksets/nemo_gym/response.py,sha256=EhrsquEP8g0bjzAThGfE6Ypgy86_nVX235NiFYtVDVM,4463
@@ -193,8 +193,8 @@ verifiers/v1/utils/scope.py,sha256=bzUEiWDOVdbj7dXOwGOlzsQ_-Xu0j9NR5TalQB-8TS8,8
 verifiers/v1/utils/score.py,sha256=493yJVMw8teCu9JxapxMFPFzI0hNUqdo0Y2nGW4kckk,6200
 verifiers/v1/utils/trace_store.py,sha256=RsCDonGR-bs-tvEstcppiJm3jFmAdIiG83fA5KtCbAw,2801
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.3.2.dev162.dist-info/METADATA,sha256=FocAQk6L9uycD720na7FpwMyDf47X7gEqG0E4lzrBUA,4161
-verifiers-0.3.2.dev162.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
-verifiers-0.3.2.dev162.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
-verifiers-0.3.2.dev162.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.3.2.dev162.dist-info/RECORD,,
+verifiers-0.3.2.dev163.dist-info/METADATA,sha256=CtlM1XVq-XCyn42L7Ouco6aCz_ijT8f-x29TYuP2dNM,4161
+verifiers-0.3.2.dev163.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
+verifiers-0.3.2.dev163.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
+verifiers-0.3.2.dev163.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.3.2.dev163.dist-info/RECORD,,