verifiers 0.2.1.dev22__py3-none-any.whl → 0.2.1.dev24__py3-none-any.whl

verifiers/v1/mcp/launch.py

@@ -263,7 +263,9 @@ async def serve(
         # `state_base`, which is universally reachable (the interception is exposed via a tunnel
         # whenever any consumer is remote). Eval-level shared servers get no per-rollout channel
         # (`state_base` is None for them).
-        state_url = f"{state_base.rstrip('/')}/state" if state_base else None
+        state_url = (
+            f"{runtime.host_url(state_base.rstrip('/'))}/state" if state_base else None
+        )
         port = await serve_in_runtime(
             server,
             runtime,
@@ -289,6 +291,8 @@ async def serve(
                 runtime, port, colocated=colocated, consumer_is_local=consumer_is_local
             )
         )
+        if not for_host and not colocated and harness_runtime is not None:
+            base = harness_runtime.host_url(base)
         yield f"{base.rstrip('/')}/mcp"
 
 
@@ -389,7 +393,8 @@ async def serve_tools(
                 urls[name] = server.url
                 logger.info("tool server '%s' (shared, external): %s", name, server.url)
                 continue
-            urls[name] = _shared_url_for_rollout(server.url, state_base, state_secret)
+            url = harness_runtime.host_url(server.url) if server.local else server.url
+            urls[name] = _shared_url_for_rollout(url, state_base, state_secret)
             # The tagged URL contains the bearer secret; log only the untagged base URL.
             logger.info("tool server '%s' (shared): %s", name, server.url)
         for toolset in toolsets:

verifiers/v1/rollout.py

@@ -158,7 +158,7 @@ class Rollout:
             async with self._serve_interception(
                 runtime, session, [*tool_servers, *([user] if user else [])]
             ) as (base_url, secret):
-                endpoint = f"{base_url}/v1"
+                endpoint = f"{runtime.host_url(base_url)}/v1"
                 async with (
                     serve_tools(
                         tool_servers,

verifiers/v1/runtimes/base.py

@@ -19,13 +19,10 @@ from verifiers.v1.utils.aio import run_shielded
 
 logger = logging.getLogger(__name__)
 
-# Ensure `uv` is available to run our PEP 723 scripts (the harness + tool servers): use it
-# if present, else bootstrap it — via pip; else via the standalone installer (curl/wget),
-# first installing curl + CA certs from the distro package manager when the image has no
-# downloader at all (bare task images, e.g. Harbor's). It installs to ~/.local/bin, which
-# we prepend to PATH so the provisioning command finds it; uv resolves each script's inline
-# deps into its own cache, isolated from the eval process. (Needs network + one of
-# uv / pip / curl / wget / apt-get / apk.)
+# Ensure the latest `uv` is available for our PEP 723 scripts: prefer pip on Python images,
+# then fall back to the standalone installer (curl/wget), installing curl + CA certs when a
+# bare image has no downloader. Both paths install to ~/.local/bin, which we prepend to PATH.
+# (Needs network + one of pip / curl / wget / apt-get / apk.)
 _INSTALL_CURL = (  # only when the image has no downloader; needs a known package manager
     "{ command -v curl >/dev/null 2>&1 || command -v wget >/dev/null 2>&1; } "
     "|| { apt-get update -qq && apt-get install -y -qq curl ca-certificates; } "
@@ -37,12 +34,8 @@ _DOWNLOAD_UV = (
 )
 _ENSURE_UV = (
     'export PATH="$HOME/.local/bin:$PATH" UV_INSTALL_DIR="$HOME/.local/bin"; '
-    # Always install the latest uv into $HOME/.local/bin (ahead of any image uv on PATH) rather
-    # than reusing whatever the image ships: base images carry wildly varying uvs (too old for the
-    # `uv sync --script` / `uv python find --script` that prepare_uv_script runs, or stale/shadowed
-    # on PATH). Installing fresh sidesteps all version probing. Falls back to pip when no downloader.
-    f"{{ {_INSTALL_CURL}; {_DOWNLOAD_UV}; }} "
-    "|| pip install -q -U uv 2>/dev/null"
+    "pip install -q -U --user uv 2>/dev/null "
+    f"|| {{ {_INSTALL_CURL}; {_DOWNLOAD_UV}; }}"
 )
 
 # The single port a self-publishing runtime (modal/prime) forwards to a public URL for a server
@@ -244,6 +237,10 @@ class Runtime(ABC):
     async def write(self, path: str, data: bytes) -> None:
         pass
 
+    def host_url(self, url: str) -> str:
+        """The URL a program inside this runtime uses to reach a host-bound `url`."""
+        return url
+
     @property
     def published_port(self) -> int | None:
         """A fixed port this runtime exposes to the outside at startup, declared up front to the

verifiers/v1/runtimes/docker.py

@@ -5,8 +5,10 @@ import contextlib
 import logging
 import shlex
 import subprocess
+import sys
 from pathlib import PurePosixPath
 from typing import Literal
+from urllib.parse import urlsplit
 
 from pydantic_config import BaseConfig
 
@@ -118,6 +120,14 @@ class DockerRuntime(Runtime):
             self.config.image,
         )
 
+    def host_url(self, url: str) -> str:
+        # Docker Desktop (macOS/Windows) runs containers in a VM, so `--network host`
+        # doesn't reach the host's loopback; `host.docker.internal` does.
+        host = urlsplit(url).hostname
+        if sys.platform != "linux" and host in ("127.0.0.1", "localhost"):
+            return url.replace(host, "host.docker.internal", 1)
+        return url
+
     async def run(self, argv: list[str], env: dict[str, str]) -> ProgramResult:
         env_args = [arg for k, v in env.items() for arg in ("--env", f"{k}={v}")]
         return await docker(

{verifiers-0.2.1.dev22.dist-info → verifiers-0.2.1.dev24.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.1.dev22
+Version: 0.2.1.dev24
 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-0.2.1.dev22.dist-info → verifiers-0.2.1.dev24.dist-info}/RECORD

@@ -209,7 +209,7 @@ verifiers/v1/legacy.py,sha256=R5vY7Q-wSY7CCyb3WGI2qJ52SwaVw0wrY5Bc-XEsNv0,20462
 verifiers/v1/loaders.py,sha256=UBiwHl76cjgw41xsz07bc9hKg1cnT1c4m05G-R1ipt4,5044
 verifiers/v1/push.py,sha256=-jZdFjPTLuyBD5H_YeUYz5KFpbFHKzIhnW3BtiY_f8o,8545
 verifiers/v1/retries.py,sha256=oZYN0S5XQjvMHLxB_06iMNAzBMjlKotTwuQtgIGUgfw,5748
-verifiers/v1/rollout.py,sha256=tIE460aYUPyAA3O_Jna5vRxJB7vUBcWr3PRXupLEHwU,10182
+verifiers/v1/rollout.py,sha256=Y069AdltwLo5Bci_2Mg723TttTq5Tajnfr3TJA6Rn2Y,10200
 verifiers/v1/scoring.py,sha256=I_mtqhTZ195hj2-CB5jzr3BY5GFKgNEvTf331f7Is4k,5740
 verifiers/v1/session.py,sha256=1ur9JZbNd3NfFSy6LNJpQYveVZXW71-S6xme54AfVnY,4796
 verifiers/v1/state.py,sha256=hYo7DkOMNyh5pcKz7Xr7j-qQN2-LOKKDTlD-KdKmW40,698
@@ -291,13 +291,13 @@ verifiers/v1/judges/reference.txt,sha256=Ej35kGXiT2uJ0LcHIeez49rCS_9uHmCQKoVdn93
 verifiers/v1/judges/rubric.py,sha256=HEWFLhTKYPO6c1uuY4FY2iKDyQnan-6DiLB5NeQpiPI,11856
 verifiers/v1/judges/rubric.txt,sha256=3KA2ZeYYdcDweJhUGopxu9EFgARhu2aWft5S5cj8Bc8,323
 verifiers/v1/mcp/__init__.py,sha256=Bn_eIjJxPXOoRqAroC_W92jDmEesUVkq-0QTUwxoV3Y,634
-verifiers/v1/mcp/launch.py,sha256=OmJfEW2eefABIcmQlrVpziRVd05Hr518G8tzj_Ta0Gc,20845
+verifiers/v1/mcp/launch.py,sha256=wt-pCIoQKsdQkYWn0v_ACkageDbp4_YzjmeVT7o60A0,21092
 verifiers/v1/mcp/server.py,sha256=MXEW6w72TkLa3iEPqrWgxQSlOBR-ostjImt5LNMtVo0,10692
 verifiers/v1/mcp/toolset.py,sha256=tSPXgTNREZyaeCnoFVnDfXxFAj_pStZMPUIgyPrY_Dk,996
 verifiers/v1/mcp/user.py,sha256=u_2KMRtNfg6WkwOCRLpR7j4ziFCUzGb1t6SomTbbu4Q,1591
 verifiers/v1/runtimes/__init__.py,sha256=L8BCIcL9InpQi3ip2-WoZKyO6FHXLnKBoYFWexEpI8Y,2138
-verifiers/v1/runtimes/base.py,sha256=jc5GoDG6oZ7gS7UKnit7zBTZyV00c8J81Mu8bj_e80g,11666
-verifiers/v1/runtimes/docker.py,sha256=yYyp8ownr4WqJbjNH5RLFVGZtDdd1_VfqUgd-rqvfog,6971
+verifiers/v1/runtimes/base.py,sha256=VlaQoOqj0-Bj2Q6nBpMtPk8Z_5caXwjdDuHzUJxcDYU,11183
+verifiers/v1/runtimes/docker.py,sha256=c96-HxwuH-aQY_mFYbFTb6TgLInUDuMTgY7z9T6DB3k,7414
 verifiers/v1/runtimes/limiters.py,sha256=LUcsvyvDkjgdDwMgx4QUduYHHPsgQJ_tgCVCV7L79Hs,2719
 verifiers/v1/runtimes/modal.py,sha256=a4NghySLu0k_YGqi59OlZdA6AqLYU96KNY2w_FExpTw,8788
 verifiers/v1/runtimes/prime.py,sha256=vnx9agHjOUxUInL99bTA9wgFAgyRdc59U6iAOmgMJDg,11547
@@ -324,8 +324,8 @@ verifiers/v1/utils/interrupt.py,sha256=F-KKhc5ndPJJfhd3SuMqyqhXhA32FhCRy5KWFJpEo
 verifiers/v1/utils/logging.py,sha256=OcMHA6NsYux3oIzjPuI95rDWmFBHNcHDjZeNIhXTX-Y,2084
 verifiers/v1/utils/memory.py,sha256=ZkIvGk6uITAH5sKon65LifKPbvZr8mJ__PVc23FZpOQ,1835
 verifiers/v1/utils/sampling.py,sha256=JczGzBn6s3wsIrSY2Hy3hE8m6NreNgsNzhSjDikQqX0,1037
-verifiers-0.2.1.dev22.dist-info/METADATA,sha256=ceFgWYA8_t864_EsZHqA1OwCTQoVbWX0QyrPS8V9ONw,5136
-verifiers-0.2.1.dev22.dist-info/WHEEL,sha256=mffPy8wBnZQn2VnJUU5jE99KsxaSfiyMHV9Yt0aLVxs,87
-verifiers-0.2.1.dev22.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
-verifiers-0.2.1.dev22.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
-verifiers-0.2.1.dev22.dist-info/RECORD,,
+verifiers-0.2.1.dev24.dist-info/METADATA,sha256=wdrMNFYferAi7vmgilNIAb9x0QWpri0ZnW8YIJMSClM,5136
+verifiers-0.2.1.dev24.dist-info/WHEEL,sha256=mffPy8wBnZQn2VnJUU5jE99KsxaSfiyMHV9Yt0aLVxs,87
+verifiers-0.2.1.dev24.dist-info/entry_points.txt,sha256=dF82JUYEFslR1AOW8LZfuBWeZeQoyX4wCRrduklg8-I,551
+verifiers-0.2.1.dev24.dist-info/licenses/LICENSE,sha256=v0RrUsdV3IDoZhrRce297IXS3xMHNJ-_LdLpFAUWb9k,1072
+verifiers-0.2.1.dev24.dist-info/RECORD,,