eval-harness

metric

rho_coherence

better
higher
kind
quality: measured against a reference or by a judge
decides
no goal
weighed by
no goal
recorded on
summeval_100

01what it measures, and how

Spearman's rho with the experts' coherence ratings across an article's 16 summaries (0 when the judge returned no coherence scores).

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

02on summeval_100

gold/summeval_100se_gemma_l_pubtext_n100_v1 · pair: 0.669se_gemma_l_n100_v1: 0.659se_gemma_l_pubtextcw_n100_v1 · pair: 0.654se_openai_l_n100_v1: 0.646se_qwen_m_n100_v1: 0.625se_anthropic_l_n100_v1: 0.606se_qwen37max_n100_v1: 0.598se_anthropic_m_n100_v1: 0.594se_qwen_s_n100_v1: 0.576se_mistral_m_n100_v1: 0.576se_unievalpub_n100_v1: 0.575se_openai_m_n100_v1: 0.574se_glm_m_n100_v1: 0.562se_glm_l_n100_v1: 0.558se_mistral_l_n100_v1: 0.552se_unieval_n100_v1: 0.552se_anthropic_s_n100_v1: 0.548se_deepseek_m_n100_v1: 0.543se_deepseek_l_n100_v1: 0.536se_gemma_m_n100_v1: 0.532se_glm_s_n100_v1: 0.524se_llama_l_n100_v1: 0.51se_deepseek_s_n100_v1: 0.509se_bartscore_n100_v1: 0.492se_gemma_s_n100_v1: 0.45se_mistral_s_n100_v1: 0.42se_llama_m_n100_v1: 0.419se_openai_s_n100_v1: 0.4se_llama_s_n100_v1: 0.375se_rougeL_n100_v1: 0.205se_rouge2_n100_v1: 0.183se_rouge1_n100_v1: 0.172se_coverage_n100_v1: 0.125se_random_n100_v1: 0.014se_length_n100_v1: -0.02-0.02better · 0.669

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 rho_coherenceitemsscored against
se_gemma_l_pubtext_n100_v1 · pair0.669100gold/summeval_100 gold
se_gemma_l_n100_v10.659100gold/summeval_100 gold
se_gemma_l_pubtextcw_n100_v1 · pair0.654100gold/summeval_100 gold
se_openai_l_n100_v10.646100gold/summeval_100 gold
se_qwen_m_n100_v10.625100gold/summeval_100 gold
se_anthropic_l_n100_v10.606100gold/summeval_100 gold
se_qwen37max_n100_v10.598100gold/summeval_100 gold
se_anthropic_m_n100_v10.594100gold/summeval_100 gold
se_qwen_s_n100_v10.576100gold/summeval_100 gold
se_mistral_m_n100_v10.576100gold/summeval_100 gold
se_unievalpub_n100_v10.575100gold/summeval_100 gold
se_openai_m_n100_v10.574100gold/summeval_100 gold
se_glm_m_n100_v10.562100gold/summeval_100 gold
se_glm_l_n100_v10.558100gold/summeval_100 gold
se_mistral_l_n100_v10.552100gold/summeval_100 gold
se_unieval_n100_v10.552100gold/summeval_100 gold
se_anthropic_s_n100_v10.548100gold/summeval_100 gold
se_deepseek_m_n100_v10.543100gold/summeval_100 gold
se_deepseek_l_n100_v10.536100gold/summeval_100 gold
se_gemma_m_n100_v10.532100gold/summeval_100 gold
se_glm_s_n100_v10.524100gold/summeval_100 gold
se_llama_l_n100_v10.510100gold/summeval_100 gold
se_deepseek_s_n100_v10.509100gold/summeval_100 gold
se_bartscore_n100_v10.492100gold/summeval_100 gold
se_gemma_s_n100_v10.450100gold/summeval_100 gold
se_mistral_s_n100_v10.420100gold/summeval_100 gold
se_llama_m_n100_v10.419100gold/summeval_100 gold
se_openai_s_n100_v10.400100gold/summeval_100 gold
se_llama_s_n100_v10.375100gold/summeval_100 gold
se_rougeL_n100_v10.205100gold/summeval_100 gold
se_rouge2_n100_v10.183100gold/summeval_100 gold
se_rouge1_n100_v10.172100gold/summeval_100 gold
se_coverage_n100_v10.125100gold/summeval_100 gold
se_random_n100_v10.014100gold/summeval_100 gold
se_length_n100_v1-0.020100gold/summeval_100 gold