eval-harness

metric

fmt_overlong

better
neutral: a description
kind
descriptive: says what the output was, never whether it was good
decides
no goal
weighed by
no goal
recorded on
cnn_dailymail_20, cnn_dailymail_200, cnn_dailymail_dm200

01what it measures, and how

1 when the output runs past 90 words where the prompt asks for 2-3 short sentences.

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

02on cnn_dailymail_20

gold/cnn_dailymail_20cnn_lead3_v1: 0.05cnn_bart_l_v1: 00better · 0.05

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 fmt_overlongitemsscored against
cnn_lead3_v10.05020gold/cnn_dailymail_20 gold
cnn_bart_l_v10.00020gold/cnn_dailymail_20 gold

03on cnn_dailymail_200

gold/cnn_dailymail_200cnn_anthropic_l_n200_v1: 0.115cnn_glm_l_n200_v1: 0.1cnn_lead3_n200_v1: 0.085cnn_lead3_n200_v1 · linux: 0.085cnn_anthropic_s_n200_v1: 0.06cnn_anthropic_m_n200_v1: 0.015cnn_deepseek_m_n200_v1: 0.01cnn_qwen_l_n200_v1: 0.005cnn_bart_l_n200_v1: 0cnn_bart_l_xsum_n200_v1: 0cnn_bart_m_n200_v1: 0cnn_bart_s_n200_v1: 0cnn_deepseek_l_n200_v1: 0cnn_deepseek_s_n200_v1: 0cnn_gemma_l_n200_v1: 0cnn_gemma_m_n200_v1: 0cnn_gemma_s_n200_v1: 0cnn_glm_m_n200_v1: 0cnn_glm_s_n200_v1: 0cnn_llama_l_n200_v1: 0cnn_llama_m_n200_v1: 0cnn_llama_s_n200_v1: 0cnn_mistral_l_n200_v1: 0cnn_mistral_m_n200_v1: 0cnn_mistral_s_n200_v1: 0cnn_openai_l_n200_v1: 0cnn_openai_m_n200_v1: 0cnn_openai_s_n200_v1: 0cnn_qwen_m_n200_v1: 0cnn_qwen_s_n200_v1: 0cnn_bart_l_n200_v1 · linux: 0cnn_bart_l_xsum_n200_v1 · linux: 0cnn_bart_m_n200_v1 · linux: 0cnn_bart_s_n200_v1 · linux: 0cnn_bart_s_n200_v1 · repeats: 0silver/cnn_dailymail_200cnn_glm_l_n200_v1 · silver: 0.1cnn_lead3_n200_v1 · silver: 0.085cnn_anthropic_s_n200_v1 · silver: 0.06cnn_anthropic_m_n200_v1 · silver: 0.015cnn_deepseek_m_n200_v1 · silver: 0.01cnn_qwen_l_n200_v1 · silver: 0.005cnn_bart_l_n200_v1 · silver: 0cnn_bart_l_xsum_n200_v1 · silver: 0cnn_bart_m_n200_v1 · silver: 0cnn_bart_s_n200_v1 · silver: 0cnn_deepseek_l_n200_v1 · silver: 0cnn_deepseek_s_n200_v1 · silver: 0cnn_gemma_l_n200_v1 · silver: 0cnn_gemma_m_n200_v1 · silver: 0cnn_gemma_s_n200_v1 · silver: 0cnn_glm_m_n200_v1 · silver: 0cnn_glm_s_n200_v1 · silver: 0cnn_llama_l_n200_v1 · silver: 0cnn_llama_m_n200_v1 · silver: 0cnn_llama_s_n200_v1 · silver: 0cnn_mistral_l_n200_v1 · silver: 0cnn_mistral_m_n200_v1 · silver: 0cnn_mistral_s_n200_v1 · silver: 0cnn_openai_l_n200_v1 · silver: 0cnn_openai_m_n200_v1 · silver: 0cnn_openai_s_n200_v1 · silver: 0cnn_qwen_m_n200_v1 · silver: 0cnn_qwen_s_n200_v1 · silver: 00better · 0.115

Not drawn: cnn_anthropic_l_n200_v1 · silver, scored against the reference it wrote (its row is in the table).

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 fmt_overlongitemsscored against
cnn_anthropic_l_n200_v10.115200gold/cnn_dailymail_200 gold
cnn_glm_l_n200_v10.100200gold/cnn_dailymail_200 gold
cnn_lead3_n200_v10.085200gold/cnn_dailymail_200 gold
cnn_lead3_n200_v1 · linux0.085200gold/cnn_dailymail_200 gold
cnn_anthropic_s_n200_v10.060200gold/cnn_dailymail_200 gold
cnn_anthropic_m_n200_v10.015200gold/cnn_dailymail_200 gold
cnn_deepseek_m_n200_v10.010200gold/cnn_dailymail_200 gold
cnn_qwen_l_n200_v10.005200gold/cnn_dailymail_200 gold
cnn_bart_l_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_bart_l_xsum_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_bart_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_bart_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_deepseek_l_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_deepseek_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_gemma_l_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_gemma_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_gemma_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_glm_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_glm_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_llama_l_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_llama_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_llama_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_mistral_l_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_mistral_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_mistral_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_openai_l_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_openai_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_openai_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_qwen_m_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_qwen_s_n200_v10.000200gold/cnn_dailymail_200 gold
cnn_bart_l_n200_v1 · linux0.000200gold/cnn_dailymail_200 gold
cnn_bart_l_xsum_n200_v1 · linux0.000200gold/cnn_dailymail_200 gold
cnn_bart_m_n200_v1 · linux0.000200gold/cnn_dailymail_200 gold
cnn_bart_s_n200_v1 · linux0.000200gold/cnn_dailymail_200 gold
cnn_bart_s_n200_v1 · repeats0.000200gold/cnn_dailymail_200 gold
cnn_anthropic_l_n200_v1 · silver0.115200silver/cnn_dailymail_200 silver
cnn_glm_l_n200_v1 · silver0.100200silver/cnn_dailymail_200 silver
cnn_lead3_n200_v1 · silver0.085200silver/cnn_dailymail_200 silver
cnn_anthropic_s_n200_v1 · silver0.060200silver/cnn_dailymail_200 silver
cnn_anthropic_m_n200_v1 · silver0.015200silver/cnn_dailymail_200 silver
cnn_deepseek_m_n200_v1 · silver0.010200silver/cnn_dailymail_200 silver
cnn_qwen_l_n200_v1 · silver0.005200silver/cnn_dailymail_200 silver
cnn_bart_l_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_bart_l_xsum_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_bart_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_bart_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_deepseek_l_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_deepseek_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_gemma_l_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_gemma_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_gemma_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_glm_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_glm_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_llama_l_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_llama_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_llama_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_mistral_l_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_mistral_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_mistral_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_openai_l_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_openai_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_openai_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_qwen_m_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver
cnn_qwen_s_n200_v1 · silver0.000200silver/cnn_dailymail_200 silver

04on cnn_dailymail_dm200

gold/cnn_dailymail_dm200cnn_bart_m_dm200_v1: 0.2cnn_lead3_dm200_v1: 0.2cnn_bart_l_dm200_v1: 0.12cnn_bart_l_xsum_dm200_v1: 0cnn_bart_s_dm200_v1: 00better · 0.2

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