test

xzrderek · xzrderek · commit d998db433f44 · 2025-10-14T13:18:06.000-07:00
diff --git a/eval_protocol/benchmarks/test_aime25.py b/eval_protocol/benchmarks/test_aime25.py
@@ -89,6 +89,7 @@ def aime2025_dataset_adapter(rows: List[Dict[str, Any]]) -> List[EvaluationRow]:
             "max_tokens": 131000,
             "extra_body": {"reasoning_effort": "low"},
             "model": "fireworks_ai/accounts/fireworks/models/gpt-oss-120b",
+            "request_timeout": 1200,  # 20 minutes Fireworks timeout
         }
     ],
     rollout_processor=SingleTurnRolloutProcessor(),

Original file line number	Diff line number	Diff line change
`@@ -89,6 +89,7 @@ def aime2025_dataset_adapter(rows: List[Dict[str, Any]]) -> List[EvaluationRow]:`
`89`	`89`	`"max_tokens": 131000,`
`90`	`90`	`"extra_body": {"reasoning_effort": "low"},`
`91`	`91`	`"model": "fireworks_ai/accounts/fireworks/models/gpt-oss-120b",`
	`92`	`+ "request_timeout": 1200, # 20 minutes Fireworks timeout`
`92`	`93`	`}`
`93`	`94`	`],`
`94`	`95`	`rollout_processor=SingleTurnRolloutProcessor(),`