verifiers 0.3.2.dev70__py3-none-any.whl → 0.3.2.dev72__py3-none-any.whl

verifiers/v1/runtimes/prime.py

@@ -295,17 +295,20 @@ class PrimeRuntime(Runtime):
 
     async def run(self, argv: list[str], env: dict[str, str]) -> ProgramResult:
         try:
-            # The shared SDK client coalesces concurrent VM job polls into batches.
-            # Rollout cancellation remains the practical execution timeout; this
-            # long SDK deadline is only a final safety bound.
-            result = await self._client.run_background_job(
+            # Poll directly so rollout cancellation owns the execution timeout.
+            job = await self._client.start_background_job(
                 self.info.id,
                 shlex.join(argv),
-                timeout=EFFECTIVELY_UNBOUNDED_SECONDS,
                 working_dir=self.config.workdir,
                 env=self.process_env(env),
-                poll_interval=1,
             )
+            delay = 0.1
+            while True:
+                result = await self._client.get_background_job(self.info.id, job)
+                if result.completed:
+                    break
+                await asyncio.sleep(delay)
+                delay = min(delay * 2, 3)
         except (
             Exception
         ) as e:  # a sandbox/API failure is one rollout's problem, not the eval's

verifiers/v1/tasksets/harbor/taskset.py

@@ -27,7 +27,7 @@ import tempfile
 import time
 from collections.abc import Iterator
 from functools import lru_cache
-from pathlib import Path
+from pathlib import Path, PurePosixPath
 from typing import Annotated
 
 from pydantic import BaseModel, ConfigDict, Field, TypeAdapter, ValidationError
@@ -250,31 +250,38 @@ class HarborTask(Task[HarborData]):
         if not self.scoring_deferred:
             trace.state.artifacts = await collect(runtime, self.data.artifacts)
 
-    async def stage_verifier(self, runtime: Runtime) -> None:
-        # An artifact under /tests must not replace the task package's verifier.
+    async def stage_verifier(self, trace: Trace, runtime: Runtime) -> None:
+        if any(
+            PurePosixPath(root).is_relative_to("/tests")
+            for root in trace.state.artifacts
+        ):
+            raise TaskError("Harbor artifacts cannot restore into /tests")
         await self.stage_tests(runtime, wipe=True)
         self.verifier_staged = True
 
     async def stage_tests(self, runtime: Runtime, wipe: bool = False) -> None:
-        """Put the task package's `tests/` in `/tests`, where `test.sh` expects it.
+        """Use a dedicated verifier image's tests, or stage the task package's tests.
 
         Raises rather than scoring stale state: a leftover reward file — planted by
         the agent or shipped in the image — must be gone before `test.sh` runs, so a
         removal that fails must not fall through to reading it.
-
-        `wipe` for a box we did not watch being built: a fresh container of the task's
-        image can ship its own `/tests`, and a leftover file there would be graded as
-        though it came from the package.
         """
-        await runtime.write(
-            "/tmp/tests.tgz", make_tar(Path(self.data.task_dir) / "tests")
-        )
-        stage = (
-            f"{'rm -rf /tests && ' if wipe else ''}"
+        # Harbor's dedicated verifier image owns the complete test suite and its
+        # dependencies. Mixing it with packaged tests can retain obsolete helpers.
+        stage = "test -f /tests/test.sh"
+        if self.data.verifier is None or self.data.verifier.image is None:
+            await runtime.write(
+                "/tmp/tests.tgz", make_tar(Path(self.data.task_dir) / "tests")
+            )
+            stage = (
+                f"{'rm -rf /tests && ' if wipe else ''}"
+                "mkdir -p /tests && tar -xzf /tmp/tests.tgz -C /tests"
+            )
+        command = (
             "rm -f /logs/verifier/reward.json /logs/verifier/reward.txt && "
-            "mkdir -p /logs/verifier /tests && tar -xzf /tmp/tests.tgz -C /tests"
+            f"mkdir -p /logs/verifier && {stage}"
         )
-        result = await runtime.run(["sh", "-c", stage], {})
+        result = await runtime.run(["sh", "-c", command], {})
         if result.exit_code:
             raise TaskError(
                 f"staging tests failed (exit {result.exit_code}): "

{verifiers-0.3.2.dev70.dist-info → verifiers-0.3.2.dev72.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev70
+Version: 0.3.2.dev72
 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.dev70.dist-info → verifiers-0.3.2.dev72.dist-info}/RECORD

@@ -144,7 +144,7 @@ verifiers/v1/runtimes/__init__.py,sha256=ppc1R_sx-aMcuJSx4t99rTSjloUXI0M6EUj676-
 verifiers/v1/runtimes/base.py,sha256=VMxYMnGE_zj4k1VseTxEtKhzOZroahRnZSfvruDenEw,17468
 verifiers/v1/runtimes/limiters.py,sha256=C6cWDD4pwyFwUY4QFke3oiMXwydSNIfSbJSIglUC2kE,3062
 verifiers/v1/runtimes/modal.py,sha256=wUNik2Wul1YISausdABIMAoSaTA0MNDwKUayKxVLwAA,13342
-verifiers/v1/runtimes/prime.py,sha256=7LIu-ciN_VT8GU6xD3W29k4ZJN1TpcL_5libDUN0VU0,19304
+verifiers/v1/runtimes/prime.py,sha256=JUj_jl1UVXHKDif41UcLXnnwX5Sdt9lQTqw14FtIl9c,19346
 verifiers/v1/runtimes/subprocess.py,sha256=QmpQ235_xrX6gowrr8557VgbydOXpLbfFq0KKGfJThk,8774
 verifiers/v1/runtimes/docker/__init__.py,sha256=JN0mI10a_BKQpRRJKX_NqStAoODAEk2dAJF9mrz69ug,20972
 verifiers/v1/runtimes/docker/egress.py,sha256=iFH_1XotlEM-OCaW7GNu3TWVEq1sqesKYweyVBX8g90,14360
@@ -157,7 +157,7 @@ verifiers/v1/serve/types.py,sha256=CjZ6PWQlzSRkBG77gXfzKHnw_S_qKUm9dZSSxVgkv1w,1
 verifiers/v1/tasksets/__init__.py,sha256=5hTKSPLXSZTgKBbeKQI7FEp06c0ZsaQfuwvrvVFLgb8,712
 verifiers/v1/tasksets/harbor/__init__.py,sha256=JxThzHFYEkuSfQbVf3pEnjordgdfe4MU-QOsSSbyovU,326
 verifiers/v1/tasksets/harbor/env.py,sha256=rZEh-1IaHpfsrwSOn7qTq7uV0K7cZZ28bHB7BEWQdnc,3622
-verifiers/v1/tasksets/harbor/taskset.py,sha256=q4Eo8vcD-3zXB7eDygUcJddmf5KpY2IBUoIlyJ_D3VU,32210
+verifiers/v1/tasksets/harbor/taskset.py,sha256=snoCQdoNBVdk9INQdfhZa8NXDXJjaidd7hU0T9cAuas,32525
 verifiers/v1/tasksets/lean/__init__.py,sha256=oyv-GbJQBvCyGwo7Ysxm2NjZW-aX4j1uDoHU0f-E7xQ,767
 verifiers/v1/tasksets/lean/scoring.py,sha256=sfzsT6MUK0zdMgAQ57_P_0CBaPHy3M7go71QMr5hDrs,10599
 verifiers/v1/tasksets/lean/taskset.py,sha256=Jax76i22G9dOAswQHiWujVbFIM_PUz5yGJI_ShR4n8Y,9238
@@ -189,8 +189,8 @@ verifiers/v1/utils/prime.py,sha256=UTYRjp9cbjNb6CVmBHda-1wWZAIIfxNOmTyNuT7_wL4,9
 verifiers/v1/utils/retries.py,sha256=Y2ZgrAjn-qNkRKZP_RVNL_7EK0iaRcZeCa404lpGYi4,5417
 verifiers/v1/utils/score.py,sha256=493yJVMw8teCu9JxapxMFPFzI0hNUqdo0Y2nGW4kckk,6200
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.3.2.dev70.dist-info/METADATA,sha256=Db5MJf31phEx5wl2P-glZjUl7InSxps2uJLXwIYy9K8,4204
-verifiers-0.3.2.dev70.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
-verifiers-0.3.2.dev70.dist-info/entry_points.txt,sha256=uqQje0TMsr7k6nXlWxlPs13Si0evb3pVCEDF6c-YL80,241
-verifiers-0.3.2.dev70.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.3.2.dev70.dist-info/RECORD,,
+verifiers-0.3.2.dev72.dist-info/METADATA,sha256=cHFZZm0kOzbpSL2Z69ikceGGWhtWO-DWCx6NlpM1gw8,4204
+verifiers-0.3.2.dev72.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+verifiers-0.3.2.dev72.dist-info/entry_points.txt,sha256=uqQje0TMsr7k6nXlWxlPs13Si0evb3pVCEDF6c-YL80,241
+verifiers-0.3.2.dev72.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.3.2.dev72.dist-info/RECORD,,