verifiers 0.2.2.dev57__tar.gz → 0.2.2.dev58__tar.gz

{verifiers-0.2.2.dev57 → verifiers-0.2.2.dev58}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev57
+Version: 0.2.2.dev58
 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.dev57 → verifiers-0.2.2.dev58}/tests/v1/fixtures/echo_tool_v1.py

@@ -1,12 +1,13 @@
 """echo (v1, MCP tool): retrieve a stamped echo from a `vf.Toolset`, then report it.
 
-The v1 tool fixture for the e2e matrix. The task declares an `EchoToolset` (`vf.Toolset`)
-with one `@vf.tool` method whose placement is CLI-tunable (`--taskset.task.tools.colocated`,
-`--taskset.task.tools.runtime.type`): it runs colocated in the harness's runtime or in its
-own runtime, and the harness must reach it wherever it lives. The tool stamps its output
-with a token the prompt never reveals, so the reward is 1.0 only if the model actually
-called the tool — trivial when the infra works, impossible when it doesn't. The tool is
-task-agnostic, so it would also serve taskset-scoped (`Taskset.tools`).
+The v1 tool fixture for the e2e matrix. The task constructs an `EchoToolset`
+(`vf.Toolset`) in `Task.toolsets`, with one `@vf.tool` method whose placement is
+CLI-tunable (`--taskset.task.tools.colocated`, `--taskset.task.tools.runtime.type`):
+it runs colocated in the harness's runtime or in its own runtime, and the harness
+must reach it wherever it lives. The tool stamps its output with a token the prompt
+never reveals, so the reward is 1.0 only if the model actually called the tool —
+trivial when the infra works, impossible when it doesn't. The tool is task-agnostic,
+so it would also serve taskset-scoped (`Taskset.toolsets`).
 """
 
 import verifiers.v1 as vf

{verifiers-0.2.2.dev57 → verifiers-0.2.2.dev58}/tests/v1/test_e2e.py

@@ -234,7 +234,7 @@ async def test_tool(run_v1, harness_runtime, tool_runtime, tmp_path):
     the harness's runtime, or its own runtime) x the harness `runtime`. The tool stamps
     its output with a token the prompt never reveals, so reward 1.0 proves the tool was
     reachable from wherever the harness runs and actually ran. Eval-wide SHARED servers
-    are a different scope (`Taskset.tools`) with their own env-server-path coverage:
+    are a different scope (`Taskset.toolsets`) with their own env-server-path coverage:
     `test_shared_tool_isolation`."""
     (trace,) = await run_v1(
         "echo-tool-v1",

{verifiers-0.2.2.dev57 → verifiers-0.2.2.dev58}/verifiers/v1/env.py

@@ -379,7 +379,7 @@ class Env(ABC, Generic[ConfigT]):
     def _requires_tunnel(self, shared: dict[str, SharedToolServer]) -> bool:
         """`requires_tunnel` over the consumers known before any rollout: role
         runtimes, live `shared` servers, and the task class's tool servers
-        (their configs read off the declared `tools` fields, no task needed)."""
+        (`Task.toolsets` is a classmethod, so no task instance is needed)."""
         task_cls = type(self.taskset).task_type()
         configs = [
             server.config for server in task_cls.toolsets(self.taskset.config.task)

{verifiers-0.2.2.dev57 → verifiers-0.2.2.dev58}/verifiers/v1/task.py

@@ -3,8 +3,8 @@
 `TaskData` is the wire half: a frozen pydantic model carrying the data which
 initializes a task instance. Rides on `trace.task.data` in `traces.jsonl`.
 
-`Task` is the behavior half: runtime prep (`setup`/`finalize`), tool declarations
-(`tools`), and scoring (`@reward`/`@metric`) methods.
+`Task` is the behavior half: runtime prep (`setup`/`finalize`), tool servers
+(`toolsets`), and scoring (`@reward`/`@metric`) methods.
 """
 
 from __future__ import annotations