verifiers 0.3.2.dev108__py3-none-any.whl → 0.3.2.dev110__py3-none-any.whl

verifiers/v1/configs/judge.py

@@ -17,7 +17,7 @@ class JudgeConfig(BaseClientConfig):
     name: str = ""
     """Reward key override for a plugged judge."""
     weight: FiniteFloat = 1.0
-    model: str = "openai/gpt-5.4-nano"
+    model: str = "openai/gpt-5.6-luna"
     sampling: SamplingConfig = SamplingConfig()
     prompt: Path | None = None
     """File whose text overrides the judge's default prompt template."""

verifiers/v1/runtimes/prime.py

@@ -1,10 +1,4 @@
-"""Remote Prime sandbox runtime.
-
-`expose` (sandbox port -> public URL) uses the SDK's native exposure (`client.expose`), so a
-host-side harness/framework can reach a tool server hosted in the sandbox. The reverse
-direction (a program in the sandbox reaching a host service) is the shared host-side
-`Tunnel` (interception.tunnel), not the runtime's concern.
-"""
+"""Remote Prime VM sandbox runtime."""
 
 import asyncio
 import base64
@@ -25,7 +19,6 @@ from pydantic import Field, model_validator
 from verifiers.v1.configs.runtime import NetworkPolicyConfig
 from verifiers.v1.errors import SandboxError
 from verifiers.v1.runtimes.base import (
-    SERVICE_PORT,
     BaseRuntimeInfo,
     ProgramResult,
     Runtime,
@@ -73,15 +66,11 @@ class PrimeConfig(NetworkPolicyConfig):
     type: Literal["prime"] = "prime"
     image: str = "python:3.11-slim"
     """Docker image to run. Any pullable ref works: on the first use of an image, the
-    platform auto-builds what the sandbox needs from it (a VM image for `vm` sandboxes,
-    ~10 minutes) and caches the result, so later sandboxes on the same ref start in
+    platform auto-builds the VM image that the sandbox needs from it (~10 minutes)
+    and caches the result, so later sandboxes on the same ref start in
     seconds."""
     workdir: str | None = None
     """Working directory override; None uses the task's workdir, or /app."""
-    vm: bool = True
-    """Run as a micro-VM rather than a container (kernel features / stronger isolation)."""
-    guaranteed: bool = False
-    """Request guaranteed (vs best-effort) capacity."""
     region: str | None = None
     """Region to provision in (None = provider-chosen)."""
     labels: list[str] = Field(default_factory=list)
@@ -106,10 +95,6 @@ class PrimeConfig(NetworkPolicyConfig):
     def _validate_egress(self) -> "PrimeConfig":
         if not self.network_restricted:
             return self
-        if not self.vm:
-            raise ValueError(
-                "Prime allow/block egress lists require a VM sandbox (vm=true)"
-            )
         if not self.allow:
             return self
         validate_egress_lists(
@@ -156,11 +141,7 @@ class PrimeRuntime(Runtime):
 
     @property
     def supports_live_processes(self) -> bool:
-        return self.config.vm
-
-    @property
-    def published_port(self) -> int | None:
-        return SERVICE_PORT
+        return True
 
     async def start(self) -> None:
         from prime_sandboxes import AsyncSandboxClient, CreateSandboxRequest
@@ -186,14 +167,7 @@ class PrimeRuntime(Runtime):
             "memory_gb": self.config.memory,
             "disk_size_gb": self.config.disk,
             "gpu_count": gpu_count,
-            # -1 is prime's convention for no lifetime limit; containers with an
-            # idle timeout must carry a finite lifetime as a safety fallback (which
-            # must exceed the idle timeout)
-            "timeout_minutes": (
-                -1
-                if self.config.vm or idle_minutes is None
-                else max(EFFECTIVELY_UNBOUNDED_SECONDS // 60, idle_minutes + 1)
-            ),
+            "timeout_minutes": -1,
             "idle_timeout_minutes": idle_minutes,
             "gpu_type": gpu_type,
             "region": self.config.region,
@@ -216,8 +190,6 @@ class PrimeRuntime(Runtime):
                                 dict.fromkeys([*BASE_LABELS, *self.config.labels])
                             ),
                             docker_image=self.config.image,
-                            vm=self.config.vm,
-                            guaranteed=self.config.guaranteed,
                             environment_vars=self.env,
                             **{k: v for k, v in options.items() if v is not None},
                         )
@@ -323,11 +295,6 @@ class PrimeRuntime(Runtime):
     async def open_process(
         self, argv: list[str], env: dict[str, str]
     ) -> RuntimeProcess:
-        if not self.config.vm:
-            raise SandboxError(
-                "persistent harness sessions on Prime require a VM sandbox; "
-                "set runtime.prime.vm=true"
-            )
         try:
             process = await self._client.open_process(
                 self.info.id,
@@ -339,22 +306,11 @@ class PrimeRuntime(Runtime):
             raise SandboxError(f"prime live process failed to start: {e}") from e
         return PrimeProcess(process)
 
-    async def expose(self, port: int) -> str | None:
-        # Publish a server hosted IN the sandbox via the SDK's native port exposure → a public
-        # HTTPS URL. Removed when the sandbox is deleted in stop(), so a tool in its own prime
-        # sandbox needs no host tunnel. Port exposure is region-gated: many regions (incl. the
-        # backend default, which lands in us-central) 400 it; `us` supports it. TODO: re-enable the
-        # prime cases in the e2e `skip_if_unexposable` guard once prime exposes ports in any region.
-        try:
-            exposed = await self._client.expose(self.info.id, port)
-        except Exception as e:  # surface prime's exposure constraints actionably
-            raise SandboxError(
-                "prime port exposure failed — port exposure isn't supported in this sandbox's "
-                "region; pin `tools.runtime.region` to a region that supports it (e.g. `us`), or "
-                f"use a colocated / docker / modal tools.runtime instead. ({e})"
-            ) from e
-        logger.info("prime: exposed sandbox port %d at %s", port, exposed.url)
-        return exposed.url.rstrip("/")
+    async def expose(self, port: int) -> str:
+        raise SandboxError(
+            "Prime VM sandboxes do not support port exposure; colocate the service "
+            "with its consumer, or use a docker or modal service runtime"
+        )
 
     async def run_background(
         self, argv: list[str], env: dict[str, str], log: str
@@ -371,7 +327,7 @@ class PrimeRuntime(Runtime):
             raise SandboxError(f"prime background launch failed: {e}") from e
 
     async def _read(self, path: str, max_bytes: int | None = None) -> bytes:
-        if max_bytes is not None and self.config.vm:
+        if max_bytes is not None:
             try:
                 # VM execute_command uses bash and returns the complete output stream.
                 result = await self._client.execute_command(
@@ -386,8 +342,6 @@ class PrimeRuntime(Runtime):
             if result.exit_code:
                 raise SandboxError(f"read {path!r}: {result.stderr.strip()[-500:]}")
             return base64.b64decode(result.stdout)
-        if max_bytes is not None:
-            return await super()._read(path, max_bytes)
         # Avoid background-job log limits and base64 overhead by downloading binary data directly.
         # The temporary file is removed on every exit, and its byte read stays off the event loop.
         target = (

verifiers/v1/task.py

@@ -104,8 +104,7 @@ class TaskData(BaseModel):
     network_allow: list[str] = Field(default_factory=lambda: ["*"])
     """Execution-time destinations requested by this task. `*` leaves the runtime
     allowlist unchanged; a concrete list replaces a wildcard or retains entries also
-    present in an existing allowlist. Prime runtimes accept host-level entries and
-    require `vm=true`."""
+    present in an existing allowlist. Prime runtimes accept host-level entries."""
     network_block: list[str] = Field(default_factory=list)
     """Execution-time destinations denied by this task and combined with runtime
     blocks. Non-empty concrete allowlists cannot be combined with blocklists. Docker

verifiers/v1/tasksets/harbor/env.py

@@ -37,7 +37,7 @@ class HarborEnv(IsolatedVerifierEnv, vf.Env[HarborEnvConfig]):
             await agents.agent.run(task)
             return
         # Resolve the verifier's box before the solve, so an impossible pairing
-        # (e.g. a restricted Prime verifier without vm=true) costs nothing
+        # (e.g. an invalid network policy) costs nothing
         # rather than a full agent run.
         self.verifier_config(task)
         await agents.agent.run(task.defer_scoring(), collect_artifacts=True)

{verifiers-0.3.2.dev108.dist-info → verifiers-0.3.2.dev110.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifiers
-Version: 0.3.2.dev108
+Version: 0.3.2.dev110
 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
@@ -36,7 +36,7 @@ Requires-Dist: numpy>=2.1.0
 Requires-Dist: openai<3.0.0,>=2.54.0
 Requires-Dist: prime-pydantic-config[toml]>=0.4.3
 Requires-Dist: prime-runs>=0.1.3
-Requires-Dist: prime-sandboxes>=0.2.39
+Requires-Dist: prime-sandboxes>=0.3.0
 Requires-Dist: prime-tunnel>=0.1.8
 Requires-Dist: pydantic>=2.12.3
 Requires-Dist: pyzmq>=27.1.0

{verifiers-0.3.2.dev108.dist-info → verifiers-0.3.2.dev110.dist-info}/RECORD

@@ -12,7 +12,7 @@ verifiers/v1/rollout.py,sha256=oka3sNdv1EbE6LatFFRMJH-5ti6FcqFuD8xa7oVUyDA,24405
 verifiers/v1/semantic.py,sha256=VQ_buw6Ya6QtRoh_fiYI5Avs5RmIQCfWH0H0bQc4aoc,4704
 verifiers/v1/session.py,sha256=7iarEDRH2epID_O8ai-1HwveXze_o8shnLw9pFMFX5o,18900
 verifiers/v1/state.py,sha256=EckF2bWp-vV4b1jYJ9sLI5xrfGuI5spIgYYwW926toI,595
-verifiers/v1/task.py,sha256=k2OImOcSohtrqmqRZS2Gv1JauOrEMPrtjSDS1o0ZTjo,12453
+verifiers/v1/task.py,sha256=7iJLXQo7JKX5jG87atIwT_HZlSto_3a7yB3w__uQQQw,12427
 verifiers/v1/taskset.py,sha256=fp2E0IEhL_Ybj9cegZwljfTmW27p_30zTWHFKw_kXE4,4374
 verifiers/v1/trace.py,sha256=uOlrvP1-C4TYtcK3S6zcCNK6UlS78hUoL-KGvJL0IsQ,32011
 verifiers/v1/types.py,sha256=PH1SxRWXOjPCgQD9aZgM18_bECvFSNQ25gkCbulmTKU,9857
@@ -47,7 +47,7 @@ verifiers/v1/configs/agent.py,sha256=PiWMNueum67vnyLLpIgxXZDb3FTQEUFqo2Z1siOOJqs
 verifiers/v1/configs/client.py,sha256=fPWtBJv5lPkqttwk0AF2ubfLdzxX2XCO0Zr-A9V4mtk,4440
 verifiers/v1/configs/env.py,sha256=-Bcdd4MEPfekVc_fnpiwsb9UDwCNdNCKrAHGoQuuXu8,7784
 verifiers/v1/configs/harness.py,sha256=ZrR-Egzfft95oMoqQ78bAq9TqP882Q68M_47Ajw9lE8,2918
-verifiers/v1/configs/judge.py,sha256=CBmGgyHXVpcBctvrFD5GlW75wpeFXVFajq_xiMaAVDk,2142
+verifiers/v1/configs/judge.py,sha256=WEwA8D4X5JSaIP3c-E2wPnvgByWeT6JRAn1d8gfiQ40,2142
 verifiers/v1/configs/retries.py,sha256=-nPlgmz7J_NZ4GOYG6oCc3thXo4La2SufBmvQGvvPGo,821
 verifiers/v1/configs/runtime.py,sha256=IBX3FxiJ2RJaiX2aKghQ15clFHvJpPUXy8GWQOpNhYM,2967
 verifiers/v1/configs/serve.py,sha256=oa_0BrAVEKboU344w0hYkSaXj8lf26t0H87gvncAo0Q,2716
@@ -145,7 +145,7 @@ verifiers/v1/runtimes/base.py,sha256=IKmrJSJltWX9M1nPUZExKluPJM4N82VsnqQGFz0ECV8
 verifiers/v1/runtimes/container.py,sha256=puQD8X9acP6GefoCNEnq40POi-nXPD-729bUjTVShfM,9554
 verifiers/v1/runtimes/limiters.py,sha256=C6cWDD4pwyFwUY4QFke3oiMXwydSNIfSbJSIglUC2kE,3062
 verifiers/v1/runtimes/modal.py,sha256=rFbxdDBi_VIiDA67XyZg-AOsc_XkV8aQWSKnTUAgCmU,17506
-verifiers/v1/runtimes/prime.py,sha256=VWzskEtMEwND6tKKhRtFM5m4Qzt001xCaI4zKqNon80,19490
+verifiers/v1/runtimes/prime.py,sha256=eeYGu60f8PSGtL6g0nJslYzrQLHo6d64JP73N3q3xTw,16982
 verifiers/v1/runtimes/subprocess.py,sha256=QmpQ235_xrX6gowrr8557VgbydOXpLbfFq0KKGfJThk,8774
 verifiers/v1/runtimes/docker/__init__.py,sha256=2wceP-8V-w0dQBEeLtoJFfCZ135DxP2rcdcUBmSWiNU,18543
 verifiers/v1/runtimes/docker/egress.py,sha256=tGwSvHNgV3-uhjshg8sZ4fVRyq8Fg7DbJl_c-lacPi4,23876
@@ -158,7 +158,7 @@ verifiers/v1/serve/server.py,sha256=i6XrY_PMAapkrP54tYX8-K6iGmOmrR7BbQ-3Qhn3w4s,
 verifiers/v1/serve/types.py,sha256=Qw5IjiJhsZz4giSXUBi9dfInBnQZ5_xiUmgH2qp8jPI,1741
 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/env.py,sha256=ATMkfwmrE_2jpOOffbjLQ91feG7TOtefBGlpwWbRq8c,3604
 verifiers/v1/tasksets/harbor/taskset.py,sha256=3xCxtVVwQSpaLfoV1SaUoNQlpIGaTmwW6PvPyYh3yv4,33780
 verifiers/v1/tasksets/harbor/toolset.py,sha256=_7HXu-uuvqzhmJ1EX6c3KWLg0g0eRrLjCq1kBsZhIwo,3159
 verifiers/v1/tasksets/lean/__init__.py,sha256=oyv-GbJQBvCyGwo7Ysxm2NjZW-aX4j1uDoHU0f-E7xQ,767
@@ -192,8 +192,8 @@ 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.dev108.dist-info/METADATA,sha256=lxM8jg5DbJZ3pMRirz1J-1ZscpIH56NZksQHizJXj5Q,4238
-verifiers-0.3.2.dev108.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
-verifiers-0.3.2.dev108.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
-verifiers-0.3.2.dev108.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.3.2.dev108.dist-info/RECORD,,
+verifiers-0.3.2.dev110.dist-info/METADATA,sha256=eD_fDZYzrtDbGvuxcBoZmYyr56UosFpv5gqh7q-F97o,4237
+verifiers-0.3.2.dev110.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+verifiers-0.3.2.dev110.dist-info/entry_points.txt,sha256=iugElcdWPKbQM7uFF0lZ8iUpHsNr17-BwEAAjJWxV3U,259
+verifiers-0.3.2.dev110.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.3.2.dev110.dist-info/RECORD,,