eval-harness

metric

rho_relevance

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' relevance ratings across an article's 16 summaries (0 when the judge returned no relevance scores).

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

02on summeval_100

gold/summeval_100se_openai_l_n100_v1: 0.591se_gemma_l_pubtext_n100_v1 · pair: 0.584se_gemma_l_n100_v1: 0.575se_openai_m_n100_v1: 0.572se_qwen37max_n100_v1: 0.569se_gemma_l_pubtextcw_n100_v1 · pair: 0.564se_qwen_s_n100_v1: 0.555se_anthropic_l_n100_v1: 0.547se_qwen_m_n100_v1: 0.539se_deepseek_l_n100_v1: 0.535se_mistral_m_n100_v1: 0.526se_deepseek_m_n100_v1: 0.526se_glm_m_n100_v1: 0.524se_gemma_m_n100_v1: 0.522se_deepseek_s_n100_v1: 0.517se_glm_s_n100_v1: 0.513se_glm_l_n100_v1: 0.512se_mistral_l_n100_v1: 0.51se_anthropic_m_n100_v1: 0.505se_anthropic_s_n100_v1: 0.505se_llama_l_n100_v1: 0.496se_mistral_s_n100_v1: 0.458se_llama_m_n100_v1: 0.444se_gemma_s_n100_v1: 0.44se_openai_s_n100_v1: 0.438se_unievalpub_n100_v1: 0.425se_unieval_n100_v1: 0.42se_llama_s_n100_v1: 0.401se_rouge1_n100_v1: 0.327se_bartscore_n100_v1: 0.324se_rouge2_n100_v1: 0.29se_coverage_n100_v1: 0.284se_rougeL_n100_v1: 0.276se_length_n100_v1: 0.124se_random_n100_v1: 0.00540.0054better · 0.591

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_relevanceitemsscored against
se_openai_l_n100_v10.591100gold/summeval_100 gold
se_gemma_l_pubtext_n100_v1 · pair0.584100gold/summeval_100 gold
se_gemma_l_n100_v10.575100gold/summeval_100 gold
se_openai_m_n100_v10.572100gold/summeval_100 gold
se_qwen37max_n100_v10.569100gold/summeval_100 gold
se_gemma_l_pubtextcw_n100_v1 · pair0.564100gold/summeval_100 gold
se_qwen_s_n100_v10.555100gold/summeval_100 gold
se_anthropic_l_n100_v10.547100gold/summeval_100 gold
se_qwen_m_n100_v10.539100gold/summeval_100 gold
se_deepseek_l_n100_v10.535100gold/summeval_100 gold
se_mistral_m_n100_v10.526100gold/summeval_100 gold
se_deepseek_m_n100_v10.526100gold/summeval_100 gold
se_glm_m_n100_v10.524100gold/summeval_100 gold
se_gemma_m_n100_v10.522100gold/summeval_100 gold
se_deepseek_s_n100_v10.517100gold/summeval_100 gold
se_glm_s_n100_v10.513100gold/summeval_100 gold
se_glm_l_n100_v10.512100gold/summeval_100 gold
se_mistral_l_n100_v10.510100gold/summeval_100 gold
se_anthropic_m_n100_v10.505100gold/summeval_100 gold
se_anthropic_s_n100_v10.505100gold/summeval_100 gold
se_llama_l_n100_v10.496100gold/summeval_100 gold
se_mistral_s_n100_v10.458100gold/summeval_100 gold
se_llama_m_n100_v10.444100gold/summeval_100 gold
se_gemma_s_n100_v10.440100gold/summeval_100 gold
se_openai_s_n100_v10.438100gold/summeval_100 gold
se_unievalpub_n100_v10.425100gold/summeval_100 gold
se_unieval_n100_v10.420100gold/summeval_100 gold
se_llama_s_n100_v10.401100gold/summeval_100 gold
se_rouge1_n100_v10.327100gold/summeval_100 gold
se_bartscore_n100_v10.324100gold/summeval_100 gold
se_rouge2_n100_v10.290100gold/summeval_100 gold
se_coverage_n100_v10.284100gold/summeval_100 gold
se_rougeL_n100_v10.276100gold/summeval_100 gold
se_length_n100_v10.124100gold/summeval_100 gold
se_random_n100_v10.005100gold/summeval_100 gold