eval-harness

metric

rho_fluency

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

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

02on summeval_100

gold/summeval_100se_glm_m_n100_v1: 0.559se_openai_m_n100_v1: 0.552se_anthropic_l_n100_v1: 0.54se_gemma_l_pubtextcw_n100_v1 · pair: 0.539se_gemma_l_pubtext_n100_v1 · pair: 0.539se_qwen_s_n100_v1: 0.539se_anthropic_s_n100_v1: 0.534se_gemma_l_n100_v1: 0.518se_qwen37max_n100_v1: 0.516se_deepseek_m_n100_v1: 0.515se_deepseek_l_n100_v1: 0.513se_openai_l_n100_v1: 0.508se_openai_s_n100_v1: 0.504se_qwen_m_n100_v1: 0.501se_anthropic_m_n100_v1: 0.495se_glm_l_n100_v1: 0.487se_deepseek_s_n100_v1: 0.485se_mistral_m_n100_v1: 0.484se_llama_l_n100_v1: 0.479se_mistral_l_n100_v1: 0.469se_gemma_m_n100_v1: 0.468se_glm_s_n100_v1: 0.464se_mistral_s_n100_v1: 0.462se_gemma_s_n100_v1: 0.45se_unievalpub_n100_v1: 0.441se_llama_m_n100_v1: 0.424se_unieval_n100_v1: 0.375se_llama_s_n100_v1: 0.341se_bartscore_n100_v1: 0.243se_rouge2_n100_v1: 0.154se_rougeL_n100_v1: 0.124se_rouge1_n100_v1: 0.11se_coverage_n100_v1: 0.046se_random_n100_v1: 0.0015se_length_n100_v1: 0.000690.00069better · 0.559

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_fluencyitemsscored against
se_glm_m_n100_v10.559100gold/summeval_100 gold
se_openai_m_n100_v10.552100gold/summeval_100 gold
se_anthropic_l_n100_v10.540100gold/summeval_100 gold
se_gemma_l_pubtextcw_n100_v1 · pair0.539100gold/summeval_100 gold
se_gemma_l_pubtext_n100_v1 · pair0.539100gold/summeval_100 gold
se_qwen_s_n100_v10.539100gold/summeval_100 gold
se_anthropic_s_n100_v10.534100gold/summeval_100 gold
se_gemma_l_n100_v10.518100gold/summeval_100 gold
se_qwen37max_n100_v10.516100gold/summeval_100 gold
se_deepseek_m_n100_v10.515100gold/summeval_100 gold
se_deepseek_l_n100_v10.513100gold/summeval_100 gold
se_openai_l_n100_v10.508100gold/summeval_100 gold
se_openai_s_n100_v10.504100gold/summeval_100 gold
se_qwen_m_n100_v10.501100gold/summeval_100 gold
se_anthropic_m_n100_v10.495100gold/summeval_100 gold
se_glm_l_n100_v10.487100gold/summeval_100 gold
se_deepseek_s_n100_v10.485100gold/summeval_100 gold
se_mistral_m_n100_v10.484100gold/summeval_100 gold
se_llama_l_n100_v10.479100gold/summeval_100 gold
se_mistral_l_n100_v10.469100gold/summeval_100 gold
se_gemma_m_n100_v10.468100gold/summeval_100 gold
se_glm_s_n100_v10.464100gold/summeval_100 gold
se_mistral_s_n100_v10.462100gold/summeval_100 gold
se_gemma_s_n100_v10.450100gold/summeval_100 gold
se_unievalpub_n100_v10.441100gold/summeval_100 gold
se_llama_m_n100_v10.424100gold/summeval_100 gold
se_unieval_n100_v10.375100gold/summeval_100 gold
se_llama_s_n100_v10.341100gold/summeval_100 gold
se_bartscore_n100_v10.243100gold/summeval_100 gold
se_rouge2_n100_v10.154100gold/summeval_100 gold
se_rougeL_n100_v10.124100gold/summeval_100 gold
se_rouge1_n100_v10.110100gold/summeval_100 gold
se_coverage_n100_v10.046100gold/summeval_100 gold
se_random_n100_v10.002100gold/summeval_100 gold
se_length_n100_v10.001100gold/summeval_100 gold