diff --git a/evalharness/config/default.yaml b/evalharness/config/default.yaml index 1620fbc..219a3f4 100644 --- a/evalharness/config/default.yaml +++ b/evalharness/config/default.yaml @@ -23,6 +23,7 @@ imo_answerbench: temperature: 1.0 gpqa_diamond: temperature: 1.0 + repeats: 3 max_tokens: 8192 mmlu: max_tokens: 8192 @@ -42,6 +43,7 @@ trivia_qa: max_tokens: 8192 humaneval: temperature: 1.0 + repeats: 3 live_code_bench: temperature: 1.0 longbench_v2: