verifiers 0.2.2.dev87__tar.gz → 0.2.2.dev88__tar.gz

{verifiers-0.2.2.dev87 → verifiers-0.2.2.dev88}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.2.dev87
+Version: 0.2.2.dev88
 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.dev87 → verifiers-0.2.2.dev88}/tests/v1/conftest.py

@@ -44,6 +44,8 @@ from verifiers.v1.configs.cli.eval import EvalConfig
 from verifiers.v1.trace import Trace
 from verifiers.v1.utils.loaders import load_environment
 
+CI_MODEL = "openai/gpt-5.6-luna"
+
 # Fixture tasksets/envs (echo-v1, echo-agentic-v1, echo-v0, echo-multi-v0) live in
 # tests/v1/fixtures, added to the path via `pythonpath` in pyproject so the v1 loader and the
 # v0 legacy bridge both resolve them by id (no install).
@@ -129,13 +131,12 @@ def _eval_config(
     runtime: dict | None = None,
     env: dict | None = None,
     pool: dict | None = None,
-    model: str | None = "deepseek/deepseek-v4-flash-0731",
     reasoning_effort: str | None = None,
 ) -> EvalConfig:
     """Build the smallest `EvalConfig` that still exercises the path, shared by the in-process
     (`run_v1`) and env-server (`run_v1_server`) fixtures. `taskset_overrides` merges onto the
     `{id: ...}` config; `runtime` places the `agent` seat's harness (an agent field, not a
-    harness one); `model` overrides the default text model (e.g. a VLM for an image task).
+    harness one).
 
     `harness=None` leaves every seat on its own story — the multi-agent case: there
     is no run-level harness, so a single-agent test's `harness` lands on the `agent`
@@ -178,7 +179,7 @@ def _eval_config(
         rich=False,
         output_dir=output_dir,
         **({"serve": {"pool": pool}} if pool else {}),
-        **({"model": model} if model else {}),
+        model=CI_MODEL,
     )
 
 
@@ -223,7 +224,7 @@ async def live_ctx():
 
     # Endpoint config only — each rollout builds and closes its own client.
     yield ModelContext(
-        model="deepseek/deepseek-v4-flash-0731",
+        model=CI_MODEL,
         client=EvalClientConfig(),
         sampling=SamplingConfig(max_tokens=2048),
     )
@@ -244,6 +245,7 @@ def run_v0():
     ) -> list[Trace]:
         config = EvalConfig(
             legacy={"id": env_id, "args": args or {}},
+            model=CI_MODEL,
             num_tasks=1,
             num_rollouts=n,
             sampling={"max_tokens": max_tokens},

{verifiers-0.2.2.dev87 → verifiers-0.2.2.dev88}/tests/v1/test_e2e.py

@@ -314,7 +314,6 @@ async def test_tool_response_image(run_v1, tmp_path):
         "tool-response-image-v1",
         harness="null",
         runtime={"type": "subprocess"},
-        model="openai/gpt-5.6-luna",
         reasoning_effort="none",
         output_dir=tmp_path,
         max_turns=4,

{verifiers-0.2.2.dev87 → verifiers-0.2.2.dev88}/tests/v1/test_envs.py

@@ -47,7 +47,7 @@ def test_eval(taskset: str):
         pytest.skip(f"{taskset} can't run a plain-CI smoke eval")
     if os.getenv("PRIME_API_KEY"):
         model = [
-            "-m", "openai/gpt-4.1-mini",
+            "-m", "openai/gpt-5.6-luna",
             "--client.base-url", "https://api.pinference.ai/api/v1",
             "--client.api-key-var", "PRIME_API_KEY",
         ]  # fmt: skip