test low concurrency

xzrderek · xzrderek · commit d231ba82c5aa · 2025-10-14T17:33:03.000-07:00
diff --git a/eval_protocol/benchmarks/test_aime25.py b/eval_protocol/benchmarks/test_aime25.py
@@ -98,7 +98,7 @@ def aime2025_dataset_adapter(rows: List[Dict[str, Any]]) -> List[EvaluationRow]:
     passed_threshold=0.8,
     num_runs=1,
     max_dataset_rows=30,
-    max_concurrent_rollouts=8,
+    max_concurrent_rollouts=2,
     mode="pointwise",
 )
 def test_aime25_pointwise(row: EvaluationRow) -> EvaluationRow:

Original file line number	Diff line number	Diff line change
`@@ -98,7 +98,7 @@ def aime2025_dataset_adapter(rows: List[Dict[str, Any]]) -> List[EvaluationRow]:`
`98`	`98`	`passed_threshold=0.8,`
`99`	`99`	`num_runs=1,`
`100`	`100`	`max_dataset_rows=30,`
`101`		`- max_concurrent_rollouts=8,`
	`101`	`+ max_concurrent_rollouts=2,`
`102`	`102`	`mode="pointwise",`
`103`	`103`	`)`
`104`	`104`	`def test_aime25_pointwise(row: EvaluationRow) -> EvaluationRow:`