verifiers 0.2.1.dev47__tar.gz → 0.2.1.dev48__tar.gz

{verifiers-0.2.1.dev47 → verifiers-0.2.1.dev48}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.1.dev47
+Version: 0.2.1.dev48
 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
@@ -35,7 +35,7 @@ Requires-Dist: numpy
 Requires-Dist: openai-agents>=0.0.7
 Requires-Dist: openai>=1.108.1
 Requires-Dist: prime-pydantic-config[toml]>=0.3.0.dev86
-Requires-Dist: prime-sandboxes>=0.2.27
+Requires-Dist: prime-sandboxes>=0.2.31
 Requires-Dist: prime-tunnel>=0.1.8
 Requires-Dist: pydantic>=2.11.9
 Requires-Dist: pyzmq>=27.1.0

{verifiers-0.2.1.dev47 → verifiers-0.2.1.dev48}/pyproject.toml

@@ -35,7 +35,7 @@ dependencies = [
     "openai>=1.108.1",
     "openai-agents>=0.0.7",
     "prime-tunnel>=0.1.8",
-    "prime-sandboxes>=0.2.27",
+    "prime-sandboxes>=0.2.31",
     "pydantic>=2.11.9",
     "requests",
     "rich",

{verifiers-0.2.1.dev47 → verifiers-0.2.1.dev48}/verifiers/v1/cli/dashboard/eval.py

@@ -41,6 +41,7 @@ _LABEL_WIDTH = len("timeouts")
 _STYLE = {
     "pending": "dim",
     "boot": "orange3",
+    "build": "dark_orange",
     "setup": "yellow",
     "running": "cyan",
     "finalize": "magenta",
@@ -51,6 +52,7 @@ _STYLE = {
 _MARK_LABEL = {
     "pending": "pending",
     "boot": "boot",
+    "build": "build",
     "setup": "setup",
     "running": "rollout",
     "finalize": "finalize",
@@ -445,6 +447,12 @@ def Rows(groups: list[list[Rollout]], now: float, runtime_type: str) -> Table:
                         stop = f"{stop} (truncated)".strip()
             else:
                 state, result, stop = rollout.phase, "", ""
+                # A boot stuck on a first-use platform image build reads differently from a
+                # normal boot — it can sit here for ~10 minutes (prime runtime only).
+                if state == Phase.BOOT and (
+                    getattr(t.runtime, "image_cached", None) is False
+                ):
+                    state = "build"
             runtime_id = t.runtime.id if t.runtime is not None else None
             runtime = f"{runtime_type}({runtime_id})" if runtime_id else runtime_type
             turns = t.num_turns

{verifiers-0.2.1.dev47 → verifiers-0.2.1.dev48}/verifiers/v1/runtimes/prime.py

@@ -37,6 +37,10 @@ MAX_LIFETIME = 24 * 60 * 60
 class PrimeConfig(BaseConfig):
     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
+    seconds."""
     workdir: str = "/app"
     network_access: bool = True
     vm: bool = False
@@ -74,7 +78,9 @@ class PrimeConfig(BaseConfig):
 
 
 class PrimeRuntimeInfo(PrimeConfig, BaseRuntimeInfo):
-    pass
+    image_cached: bool | None = None
+    """Whether the platform already had the image at create (None until then). False means
+    a first-use auto-build ran while this sandbox waited to start."""
 
 
 class PrimeRuntime(Runtime):
@@ -134,6 +140,19 @@ class PrimeRuntime(Runtime):
                     )
                 )
             self.info.id = sandbox.id
+            # The create response says whether the platform already has the image:
+            # `pending_image_build_id` set means a first-use auto-build is running and the
+            # sandbox stays PENDING until it finishes (`wait_for_creation` gives that phase
+            # its own budget, separate from the normal boot attempts).
+            self.info.image_cached = sandbox.pending_image_build_id is None
+            if not self.info.image_cached:
+                logger.warning(
+                    "prime: image %s isn't cached on the platform - auto-building it "
+                    "(sandbox %s waits for the build; first use of an image can take "
+                    "~10 minutes, later runs start in seconds)",
+                    self.config.image,
+                    self.info.id,
+                )
             await self._client.wait_for_creation(self.info.id)
             logger.info(
                 "prime: sandbox %s up (image=%s)", self.info.id, self.config.image