eval-harness

metric

tie_rate

better
neutral: a description
kind
descriptive: says what the output was, never whether it was good
decides
no goal
weighed by
no goal
recorded on
summeval_100

01what it measures, and how

Share of the 120 summary pairs the judge scored equal, averaged over the dimensions it returned: how often it could not tell two summaries apart.

A project explains its own metric at length with METRIC_DESCRIPTIONS = {"tie_rate": {"summary": "…", "details": "…"}}.

02on summeval_100

gold/summeval_100se_gemma_s_n100_v1: 0.498se_glm_m_n100_v1: 0.416se_gemma_l_pubtext_n100_v1 · pair: 0.416se_gemma_l_pubtextcw_n100_v1 · pair: 0.415se_gemma_l_n100_v1: 0.405se_glm_l_n100_v1: 0.399se_gemma_m_n100_v1: 0.391se_anthropic_l_n100_v1: 0.39se_llama_s_n100_v1: 0.382se_mistral_s_n100_v1: 0.374se_openai_m_n100_v1: 0.374se_openai_s_n100_v1: 0.369se_openai_l_n100_v1: 0.366se_llama_l_n100_v1: 0.363se_glm_s_n100_v1: 0.357se_deepseek_m_n100_v1: 0.356se_qwen_s_n100_v1: 0.356se_llama_m_n100_v1: 0.354se_anthropic_m_n100_v1: 0.353se_qwen37max_n100_v1: 0.348se_anthropic_s_n100_v1: 0.343se_deepseek_s_n100_v1: 0.331se_qwen_m_n100_v1: 0.331se_deepseek_l_n100_v1: 0.297se_mistral_l_n100_v1: 0.295se_mistral_m_n100_v1: 0.279se_random_n100_v1: 0.201se_length_n100_v1: 0.026se_coverage_n100_v1: 0.015se_rouge1_n100_v1: 0.013se_rouge2_n100_v1: 0.012se_rougeL_n100_v1: 0.0095se_unievalpub_n100_v1: 0.0055se_unieval_n100_v1: 0.0052se_bartscore_n100_v1: 0.00520.0052better · 0.498

scored against silverscored by a judgewrote the reference it is scored againstscored against a label set, not a reference setunmarked rows are scored against gold

armmean tie_rateitemsscored against
se_gemma_s_n100_v10.498100gold/summeval_100 gold
se_glm_m_n100_v10.416100gold/summeval_100 gold
se_gemma_l_pubtext_n100_v1 · pair0.416100gold/summeval_100 gold
se_gemma_l_pubtextcw_n100_v1 · pair0.415100gold/summeval_100 gold
se_gemma_l_n100_v10.405100gold/summeval_100 gold
se_glm_l_n100_v10.399100gold/summeval_100 gold
se_gemma_m_n100_v10.391100gold/summeval_100 gold
se_anthropic_l_n100_v10.390100gold/summeval_100 gold
se_llama_s_n100_v10.382100gold/summeval_100 gold
se_mistral_s_n100_v10.374100gold/summeval_100 gold
se_openai_m_n100_v10.374100gold/summeval_100 gold
se_openai_s_n100_v10.369100gold/summeval_100 gold
se_openai_l_n100_v10.366100gold/summeval_100 gold
se_llama_l_n100_v10.363100gold/summeval_100 gold
se_glm_s_n100_v10.357100gold/summeval_100 gold
se_deepseek_m_n100_v10.356100gold/summeval_100 gold
se_qwen_s_n100_v10.35699gold/summeval_100 gold
se_llama_m_n100_v10.354100gold/summeval_100 gold
se_anthropic_m_n100_v10.353100gold/summeval_100 gold
se_qwen37max_n100_v10.348100gold/summeval_100 gold
se_anthropic_s_n100_v10.343100gold/summeval_100 gold
se_deepseek_s_n100_v10.331100gold/summeval_100 gold
se_qwen_m_n100_v10.331100gold/summeval_100 gold
se_deepseek_l_n100_v10.297100gold/summeval_100 gold
se_mistral_l_n100_v10.295100gold/summeval_100 gold
se_mistral_m_n100_v10.279100gold/summeval_100 gold
se_random_n100_v10.201100gold/summeval_100 gold
se_length_n100_v10.026100gold/summeval_100 gold
se_coverage_n100_v10.015100gold/summeval_100 gold
se_rouge1_n100_v10.013100gold/summeval_100 gold
se_rouge2_n100_v10.012100gold/summeval_100 gold
se_rougeL_n100_v10.009100gold/summeval_100 gold
se_unievalpub_n100_v10.005100gold/summeval_100 gold
se_unieval_n100_v10.005100gold/summeval_100 gold
se_bartscore_n100_v10.005100gold/summeval_100 gold