eval-harness

metric

tau_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

Kendall's tau-b 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 = {"tau_consistency": {"summary": "…", "details": "…"}}.

02on summeval_100

gold/summeval_100se_openai_l_n100_v1: 0.635se_qwen_s_n100_v1: 0.626se_gemma_l_pubtext_n100_v1 · pair: 0.626se_anthropic_l_n100_v1: 0.621se_openai_m_n100_v1: 0.618se_glm_m_n100_v1: 0.614se_gemma_l_n100_v1: 0.614se_qwen37max_n100_v1: 0.611se_gemma_l_pubtextcw_n100_v1 · pair: 0.61se_glm_l_n100_v1: 0.596se_qwen_m_n100_v1: 0.587se_gemma_m_n100_v1: 0.576se_deepseek_m_n100_v1: 0.552se_anthropic_s_n100_v1: 0.545se_anthropic_m_n100_v1: 0.532se_mistral_s_n100_v1: 0.53se_llama_l_n100_v1: 0.517se_deepseek_s_n100_v1: 0.508se_llama_m_n100_v1: 0.503se_mistral_m_n100_v1: 0.494se_deepseek_l_n100_v1: 0.491se_mistral_l_n100_v1: 0.489se_gemma_s_n100_v1: 0.486se_glm_s_n100_v1: 0.468se_openai_s_n100_v1: 0.465se_llama_s_n100_v1: 0.449se_unievalpub_n100_v1: 0.356se_unieval_n100_v1: 0.338se_bartscore_n100_v1: 0.198se_rouge2_n100_v1: 0.148se_rouge1_n100_v1: 0.122se_rougeL_n100_v1: 0.109se_length_n100_v1: 0.071se_coverage_n100_v1: 0.056se_random_n100_v1: -0.0073-0.0073better · 0.635

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 tau_consistencyitemsscored against
se_openai_l_n100_v10.635100gold/summeval_100 gold
se_qwen_s_n100_v10.626100gold/summeval_100 gold
se_gemma_l_pubtext_n100_v1 · pair0.626100gold/summeval_100 gold
se_anthropic_l_n100_v10.621100gold/summeval_100 gold
se_openai_m_n100_v10.618100gold/summeval_100 gold
se_glm_m_n100_v10.614100gold/summeval_100 gold
se_gemma_l_n100_v10.614100gold/summeval_100 gold
se_qwen37max_n100_v10.611100gold/summeval_100 gold
se_gemma_l_pubtextcw_n100_v1 · pair0.610100gold/summeval_100 gold
se_glm_l_n100_v10.596100gold/summeval_100 gold
se_qwen_m_n100_v10.587100gold/summeval_100 gold
se_gemma_m_n100_v10.576100gold/summeval_100 gold
se_deepseek_m_n100_v10.552100gold/summeval_100 gold
se_anthropic_s_n100_v10.545100gold/summeval_100 gold
se_anthropic_m_n100_v10.532100gold/summeval_100 gold
se_mistral_s_n100_v10.530100gold/summeval_100 gold
se_llama_l_n100_v10.517100gold/summeval_100 gold
se_deepseek_s_n100_v10.508100gold/summeval_100 gold
se_llama_m_n100_v10.503100gold/summeval_100 gold
se_mistral_m_n100_v10.494100gold/summeval_100 gold
se_deepseek_l_n100_v10.491100gold/summeval_100 gold
se_mistral_l_n100_v10.489100gold/summeval_100 gold
se_gemma_s_n100_v10.486100gold/summeval_100 gold
se_glm_s_n100_v10.468100gold/summeval_100 gold
se_openai_s_n100_v10.465100gold/summeval_100 gold
se_llama_s_n100_v10.449100gold/summeval_100 gold
se_unievalpub_n100_v10.356100gold/summeval_100 gold
se_unieval_n100_v10.338100gold/summeval_100 gold
se_bartscore_n100_v10.198100gold/summeval_100 gold
se_rouge2_n100_v10.148100gold/summeval_100 gold
se_rouge1_n100_v10.122100gold/summeval_100 gold
se_rougeL_n100_v10.109100gold/summeval_100 gold
se_length_n100_v10.071100gold/summeval_100 gold
se_coverage_n100_v10.056100gold/summeval_100 gold
se_random_n100_v1-0.007100gold/summeval_100 gold