verifiers 0.2.1.dev39__tar.gz → 0.2.1.dev41__tar.gz

{verifiers-0.2.1.dev39 → verifiers-0.2.1.dev41}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: verifiers
-Version: 0.2.1.dev39
+Version: 0.2.1.dev41
 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.dev39 → verifiers-0.2.1.dev41}/verifiers/v1/gepa/config.py

@@ -21,13 +21,14 @@ from verifiers.v1.types import SamplingConfig
 class GEPAConfig(EnvConfig):
     """The GEPA run plus its environment. `model` runs the rollouts under optimization;
     `reflection_model` (defaults to `model`) proposes new system prompts from the reflective
-    dataset. No default for `model` — a GEPA run can spend a large eval budget, so pick it
-    explicitly rather than silently optimizing (and spending) against a fallback."""
+    dataset. `model` defaults to the same id as `EvalConfig`."""
 
     uuid: str = Field(default_factory=lambda: str(uuid4()), exclude=True)
     """Auto-generated run id — the leaf of the output dir, so runs never overwrite.
     Excluded from the saved config so re-running `@ config.toml` lands in a fresh dir."""
-    model: str = Field(..., validation_alias=AliasChoices("model", "m"))
+    model: str = Field(
+        "deepseek/deepseek-v4-flash", validation_alias=AliasChoices("model", "m")
+    )
     """Model id for rollouts under optimization."""
     client: EvalClientConfig = EvalClientConfig()
     sampling: SamplingConfig = SamplingConfig()
@@ -48,14 +49,10 @@ class GEPAConfig(EnvConfig):
     """Seed for GEPA's optimizer (candidate selection / minibatch sampling). Task shuffling
     uses a fixed seed, matching eval — so this doesn't change the train/val split."""
 
-    max_metric_calls: int = Field(
-        500, validation_alias=AliasChoices("max_metric_calls", "B")
-    )
+    max_total_rollouts: int = Field(500)
     """Total rollouts GEPA may spend across the whole optimization run."""
     reflection_minibatch_size: int = 3
     """Train tasks sampled per reflection step."""
-    perfect_score: float | None = None
-    """Skip reflecting on a minibatch that already scores this well."""
     reflection_columns: list[str] = Field(default_factory=list)
     """Extra per-trace fields (from `trace.info`, else `task`) to surface to the teacher LM."""
     initial_prompt: str | None = None

{verifiers-0.2.1.dev39 → verifiers-0.2.1.dev41}/verifiers/v1/gepa/runner.py

@@ -101,16 +101,14 @@ def run_gepa(env: Environment, config: GEPAConfig) -> GEPAResult:
                 valset=[task.data.idx for task in val_tasks],
                 adapter=adapter,
                 reflection_lm=reflection_lm,
-                max_metric_calls=config.max_metric_calls,
+                max_metric_calls=config.max_total_rollouts,
                 reflection_minibatch_size=config.reflection_minibatch_size,
                 run_dir=str(run_dir) if run_dir is not None else None,
                 seed=config.seed,
                 display_progress_bar=False,
-                skip_perfect_score=config.perfect_score is not None,
+                skip_perfect_score=False,
                 logger=_GEPALog(),
             )
-            if config.perfect_score is not None:
-                optimize_kwargs["perfect_score"] = config.perfect_score
             return optimize(**optimize_kwargs)
         finally:
             loop.run_until_complete(serving.__aexit__(None, None, None))

{verifiers-0.2.1.dev39 → verifiers-0.2.1.dev41}/verifiers/v1/tasksets/harbor/taskset.py

@@ -236,14 +236,33 @@ def resolve_image(
     return None
 
 
+def size_to_mb(size: str | int | float) -> float:
+    """A Harbor size in MB, from either schema: current integer-MB fields or the
+    legacy schema-1.0 size strings ("8G", "512M", "64K")."""
+    if not isinstance(size, str):
+        return float(size)
+    scale = {"G": 1024.0, "M": 1.0, "K": 1 / 1024}.get(size.strip().upper()[-1:])
+    if scale is None:
+        raise ValueError(
+            f"invalid Harbor size {size!r}: expected a number of MB or a "
+            "'<number>[G|M|K]' string"
+        )
+    return float(size.strip()[:-1]) * scale
+
+
 def parse_resources(env: dict, multiplier: float = 1.0) -> TaskResources:
-    # Harbor reports memory and storage in MB; TaskResources stores GB. A zero
-    # GPU count means no GPU request rather than the string "0".
+    # Harbor's current schema reports memory/storage as integer-MB fields; the
+    # legacy schema 1.0 used size strings under `memory`/`storage`, which Harbor
+    # still migrates (datasets like senior-swe-bench are authored against it).
+    # TaskResources stores GB. A zero GPU count means no GPU request rather than
+    # the string "0".
+    memory = env.get("memory_mb") or env.get("memory")
+    disk = env.get("storage_mb") or env.get("storage")
     return TaskResources(
         cpu=env["cpus"] * multiplier if env.get("cpus") else None,
-        memory=env["memory_mb"] / 1024 * multiplier if env.get("memory_mb") else None,
+        memory=size_to_mb(memory) / 1024 * multiplier if memory else None,
         gpu=str(env["gpus"]) if env.get("gpus") else None,
-        disk=env["storage_mb"] / 1024 * multiplier if env.get("storage_mb") else None,
+        disk=size_to_mb(disk) / 1024 * multiplier if disk else None,
     )