verifiers 0.2.2.dev44__py3-none-any.whl → 0.2.2.dev46__py3-none-any.whl

verifiers/v1/runtimes/prime.py

@@ -223,21 +223,14 @@ class PrimeRuntime(Runtime):
 
     async def run(self, argv: list[str], env: dict[str, str]) -> ProgramResult:
         try:
-            # Poll directly so the rollout stage owns the execution timeout; the SDK helper
-            # otherwise imposes its own 15-minute limit.
-            job = await self._client.start_background_job(
+            result = await self._client.run_background_job(
                 self.info.id,
                 shlex.join(argv),
+                timeout=MAX_LIFETIME,
                 working_dir=self.config.workdir,
                 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/serve/types.py

@@ -1,6 +1,6 @@
 from typing import ClassVar
 
-from pydantic import BaseModel, Field, field_serializer, model_validator
+from pydantic import BaseModel, Field, SerializeAsAny, model_validator
 
 from verifiers.v1.clients.config import ClientConfig
 from verifiers.v1.episode import WireEpisode
@@ -65,15 +65,11 @@ class RunRequest(BaseRequest):
 
 
 class RunResponse(BaseResponse):
-    episode: WireEpisode | None = None
+    episode: SerializeAsAny[WireEpisode] | None = None
     """The rollout's episode — its standing (`id`/`env`/`errors`, carrying
     episode-level errors even when no trace minted) inlined next to its flat,
     self-contained traces; task-specific data preserved in `model_extra`."""
 
-    @field_serializer("episode")
-    def _ser_episode(self, episode: "WireEpisode | None") -> dict | None:
-        return episode.model_dump() if episode is not None else None
-
 
 class RunGroupRequest(BaseRequest):
     """Legacy (v0) route: group-scored v0 envs run a task's n rollouts together."""
@@ -87,8 +83,4 @@ class RunGroupRequest(BaseRequest):
 
 
 class RunGroupResponse(BaseResponse):
-    traces: list[WireTrace] | None = None
-
-    @field_serializer("traces")
-    def _ser_traces(self, traces: "list[WireTrace] | None") -> list[dict] | None:
-        return [t.model_dump() for t in traces] if traces is not None else None
+    traces: list[SerializeAsAny[WireTrace]] | None = None

{verifiers-0.2.2.dev44.dist-info → verifiers-0.2.2.dev46.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev44
+Version: 0.2.2.dev46
 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-0.2.2.dev44.dist-info → verifiers-0.2.2.dev46.dist-info}/RECORD

@@ -297,7 +297,7 @@ verifiers/v1/runtimes/__init__.py,sha256=TqNtJVahzQuLS4fgCVeT9xdOzTkp0hRAPOkl7eF
 verifiers/v1/runtimes/base.py,sha256=sTCiLo51yAU_T-Voc3BTH-FuzyEM0cGY05VzFE4DmYc,14720
 verifiers/v1/runtimes/limiters.py,sha256=ZjjYUFEt20yH8ggL5uU9w9U500AOB44qSUTQ7bYyh0c,2730
 verifiers/v1/runtimes/modal.py,sha256=FnocoJd9lQVQS3V3CoBon7KpKXm1bKWWeU2FAkq2bKU,8825
-verifiers/v1/runtimes/prime.py,sha256=wldpcZnKZHNRldpoubeMgHdl-ZlKWOyeulFu4nHNKwc,15110
+verifiers/v1/runtimes/prime.py,sha256=zMMoB85uqwXjRTTyb3q5C7Xs4VmHagB8oI92Kuc-E1E,14755
 verifiers/v1/runtimes/subprocess.py,sha256=dOqYZycTO5wmXqg0Lg1h3l_CPh6Zok4PmTFcc-uE4WY,5989
 verifiers/v1/runtimes/docker/__init__.py,sha256=eC5hjfyPwvwEKaeZUeP9EN8tJkrl0HUxGMzjiK4tK4Q,14555
 verifiers/v1/runtimes/docker/egress.py,sha256=37UXgNX9gnoLxjMg4KZaJIGoApCAM6MK22eJGNoMi38,14514
@@ -305,7 +305,7 @@ verifiers/v1/serve/__init__.py,sha256=sr_zkDzAHnOcQm-QRm8xwJBmCMFD-xhKIFPaDZvMbU
 verifiers/v1/serve/client.py,sha256=amUhf4cFJ1xCb8kw_AoVviwN543Wknx8LzImq1ZpaiQ,7057
 verifiers/v1/serve/pool.py,sha256=e9Nc3nQKEyV9YqyKHl_RuqQ84zWVa8z2J1rc3rfCadM,14828
 verifiers/v1/serve/server.py,sha256=VBOD3Gzq4p0ZD_8xgtG039i3JaefhOH0MmR6cxG56iY,9705
-verifiers/v1/serve/types.py,sha256=AHFbt8C-Rzlee34RDPLQ3lPiYDLkUm5eFMnRQINiO6s,3016
+verifiers/v1/serve/types.py,sha256=y8Cf9wKOZRKDXcThFWVygTFHITJofOinGiIzwjQGCUE,2674
 verifiers/v1/tasksets/__init__.py,sha256=b62O4WDNbGfJH_4NjYmznn5otWovdfvUgJZyoIZyLh8,596
 verifiers/v1/tasksets/harbor/__init__.py,sha256=wwVGA8N7E2Q2qA5WGHuLRrn4DgG5oaRHrCS2RW3NXVY,195
 verifiers/v1/tasksets/harbor/taskset.py,sha256=aPgdVExCtt09t2YJrsK4UmlpQwAp5-htXa7arC1bX68,13877
@@ -329,8 +329,8 @@ verifiers/v1/utils/logging.py,sha256=OcMHA6NsYux3oIzjPuI95rDWmFBHNcHDjZeNIhXTX-Y
 verifiers/v1/utils/memory.py,sha256=ZkIvGk6uITAH5sKon65LifKPbvZr8mJ__PVc23FZpOQ,1835
 verifiers/v1/utils/sampling.py,sha256=JczGzBn6s3wsIrSY2Hy3hE8m6NreNgsNzhSjDikQqX0,1037
 verifiers/v1/utils/version.py,sha256=-obEo_-l9-D8FLef4hYxncOe-uJpxrM1g2Hig_37Sgs,1607
-verifiers-0.2.2.dev44.dist-info/METADATA,sha256=xAsUe3hwzJqF-nbNTppTvycNvcRTi_cnhIq8oFj9adQ,4545
-verifiers-0.2.2.dev44.dist-info/WHEEL,sha256=lCkmxWfQsSc9CfIClYeavTdQeEX2toPqufh9gI35EQA,87
-verifiers-0.2.2.dev44.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
-verifiers-0.2.2.dev44.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.2.2.dev44.dist-info/RECORD,,
+verifiers-0.2.2.dev46.dist-info/METADATA,sha256=W6WgwXWR8WefVXSJvKQp14Cw8-R1LKUxyIqEVn33deI,4545
+verifiers-0.2.2.dev46.dist-info/WHEEL,sha256=lCkmxWfQsSc9CfIClYeavTdQeEX2toPqufh9gI35EQA,87
+verifiers-0.2.2.dev46.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
+verifiers-0.2.2.dev46.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.2.2.dev46.dist-info/RECORD,,