eval-harness

metric

rho_consistency

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

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

02on summeval_100

gold/summeval_100se_openai_l_n100_v1: 0.663se_anthropic_l_n100_v1: 0.654se_qwen_s_n100_v1: 0.654se_gemma_l_pubtext_n100_v1 · pair: 0.651se_openai_m_n100_v1: 0.65se_qwen37max_n100_v1: 0.641se_glm_m_n100_v1: 0.64se_gemma_l_n100_v1: 0.638se_gemma_l_pubtextcw_n100_v1 · pair: 0.636se_glm_l_n100_v1: 0.629se_qwen_m_n100_v1: 0.619se_gemma_m_n100_v1: 0.604se_deepseek_m_n100_v1: 0.585se_anthropic_s_n100_v1: 0.578se_mistral_s_n100_v1: 0.565se_anthropic_m_n100_v1: 0.563se_llama_l_n100_v1: 0.547se_deepseek_s_n100_v1: 0.546se_mistral_m_n100_v1: 0.538se_llama_m_n100_v1: 0.535se_deepseek_l_n100_v1: 0.535se_mistral_l_n100_v1: 0.526se_gemma_s_n100_v1: 0.509se_glm_s_n100_v1: 0.507se_openai_s_n100_v1: 0.503se_llama_s_n100_v1: 0.474se_unievalpub_n100_v1: 0.428se_unieval_n100_v1: 0.408se_bartscore_n100_v1: 0.24se_rouge2_n100_v1: 0.178se_rouge1_n100_v1: 0.15se_rougeL_n100_v1: 0.132se_length_n100_v1: 0.085se_coverage_n100_v1: 0.066se_random_n100_v1: -0.0095-0.0095better · 0.663

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_consistencyitemsscored against
se_openai_l_n100_v10.663100gold/summeval_100 gold
se_anthropic_l_n100_v10.654100gold/summeval_100 gold
se_qwen_s_n100_v10.654100gold/summeval_100 gold
se_gemma_l_pubtext_n100_v1 · pair0.651100gold/summeval_100 gold
se_openai_m_n100_v10.650100gold/summeval_100 gold
se_qwen37max_n100_v10.641100gold/summeval_100 gold
se_glm_m_n100_v10.640100gold/summeval_100 gold
se_gemma_l_n100_v10.638100gold/summeval_100 gold
se_gemma_l_pubtextcw_n100_v1 · pair0.636100gold/summeval_100 gold
se_glm_l_n100_v10.629100gold/summeval_100 gold
se_qwen_m_n100_v10.619100gold/summeval_100 gold
se_gemma_m_n100_v10.604100gold/summeval_100 gold
se_deepseek_m_n100_v10.585100gold/summeval_100 gold
se_anthropic_s_n100_v10.578100gold/summeval_100 gold
se_mistral_s_n100_v10.565100gold/summeval_100 gold
se_anthropic_m_n100_v10.563100gold/summeval_100 gold
se_llama_l_n100_v10.547100gold/summeval_100 gold
se_deepseek_s_n100_v10.546100gold/summeval_100 gold
se_mistral_m_n100_v10.538100gold/summeval_100 gold
se_llama_m_n100_v10.535100gold/summeval_100 gold
se_deepseek_l_n100_v10.535100gold/summeval_100 gold
se_mistral_l_n100_v10.526100gold/summeval_100 gold
se_gemma_s_n100_v10.509100gold/summeval_100 gold
se_glm_s_n100_v10.507100gold/summeval_100 gold
se_openai_s_n100_v10.503100gold/summeval_100 gold
se_llama_s_n100_v10.474100gold/summeval_100 gold
se_unievalpub_n100_v10.428100gold/summeval_100 gold
se_unieval_n100_v10.408100gold/summeval_100 gold
se_bartscore_n100_v10.240100gold/summeval_100 gold
se_rouge2_n100_v10.178100gold/summeval_100 gold
se_rouge1_n100_v10.150100gold/summeval_100 gold
se_rougeL_n100_v10.132100gold/summeval_100 gold
se_length_n100_v10.085100gold/summeval_100 gold
se_coverage_n100_v10.066100gold/summeval_100 gold
se_random_n100_v1-0.010100gold/summeval_100 gold