eval-harness

metric

fmt_verbose

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
ag_news_200, dbpedia_280

01what it measures, and how

1 when the reply ran past two words, where the prompt asks for the label alone.

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

02on ag_news_200

gold/ag_news_200ag_glm_l_n200_v1: 0.09ag_anthropic_l_n200_v1: 0ag_anthropic_m_n200_v1: 0ag_anthropic_s_n200_v1: 0ag_bart_mnli_n200_v1: 0ag_bert_base_fy_n200_v1: 0ag_bert_base_ta_n200_v1: 0ag_bert_mini_n200_v1: 0ag_constant_n200_v1: 0ag_deepseek_l_n200_v1: 0ag_deepseek_m_n200_v1: 0ag_deepseek_s_n200_v1: 0ag_gemma_l_n200_v1: 0ag_gemma_m_n200_v1: 0ag_gemma_s_n200_v1: 0ag_glm_m_n200_v1: 0ag_glm_s_n200_v1: 0ag_keyword_n200_v1: 0ag_llama_l_n200_v1: 0ag_llama_m_n200_v1: 0ag_llama_s_n200_v1: 0ag_mistral_l_n200_v1: 0ag_mistral_m_n200_v1: 0ag_mistral_s_n200_v1: 0ag_openai_l_n200_v1: 0ag_openai_m_n200_v1: 0ag_openai_s_n200_v1: 0ag_qwen_l_n200_v1: 0ag_qwen_m_n200_v1: 0ag_qwen_s_n200_v1: 0ag_bert_base_fy_n200_v1 · linux: 0ag_bert_base_ta_n200_v1 · linux: 0ag_bert_mini_n200_v1 · linux: 0gold/ag_news_200_adj1ag_glm_l_n200_v1 · adjudicated: 0.09ag_anthropic_l_n200_v1 · adjudicated: 0ag_anthropic_m_n200_v1 · adjudicated: 0ag_anthropic_s_n200_v1 · adjudicated: 0ag_bart_mnli_n200_v1 · adjudicated: 0ag_bert_base_fy_n200_v1 · adjudicated: 0ag_bert_base_ta_n200_v1 · adjudicated: 0ag_bert_mini_n200_v1 · adjudicated: 0ag_constant_n200_v1 · adjudicated: 0ag_deepseek_l_n200_v1 · adjudicated: 0ag_deepseek_m_n200_v1 · adjudicated: 0ag_deepseek_s_n200_v1 · adjudicated: 0ag_gemma_l_n200_v1 · adjudicated: 0ag_gemma_m_n200_v1 · adjudicated: 0ag_gemma_s_n200_v1 · adjudicated: 0ag_glm_m_n200_v1 · adjudicated: 0ag_glm_s_n200_v1 · adjudicated: 0ag_keyword_n200_v1 · adjudicated: 0ag_llama_l_n200_v1 · adjudicated: 0ag_llama_m_n200_v1 · adjudicated: 0ag_llama_s_n200_v1 · adjudicated: 0ag_mistral_l_n200_v1 · adjudicated: 0ag_mistral_m_n200_v1 · adjudicated: 0ag_mistral_s_n200_v1 · adjudicated: 0ag_openai_l_n200_v1 · adjudicated: 0ag_openai_m_n200_v1 · adjudicated: 0ag_openai_s_n200_v1 · adjudicated: 0ag_qwen_l_n200_v1 · adjudicated: 0ag_qwen_m_n200_v1 · adjudicated: 0ag_qwen_s_n200_v1 · adjudicated: 00better · 0.09

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_verboseitemsscored against
ag_glm_l_n200_v10.090200gold/ag_news_200 gold
ag_anthropic_l_n200_v10.000200gold/ag_news_200 gold
ag_anthropic_m_n200_v10.000200gold/ag_news_200 gold
ag_anthropic_s_n200_v10.000200gold/ag_news_200 gold
ag_bart_mnli_n200_v10.000200gold/ag_news_200 gold
ag_bert_base_fy_n200_v10.000200gold/ag_news_200 gold
ag_bert_base_ta_n200_v10.000200gold/ag_news_200 gold
ag_bert_mini_n200_v10.000200gold/ag_news_200 gold
ag_constant_n200_v10.000200gold/ag_news_200 gold
ag_deepseek_l_n200_v10.000200gold/ag_news_200 gold
ag_deepseek_m_n200_v10.000200gold/ag_news_200 gold
ag_deepseek_s_n200_v10.000200gold/ag_news_200 gold
ag_gemma_l_n200_v10.000200gold/ag_news_200 gold
ag_gemma_m_n200_v10.000200gold/ag_news_200 gold
ag_gemma_s_n200_v10.000200gold/ag_news_200 gold
ag_glm_m_n200_v10.000200gold/ag_news_200 gold
ag_glm_s_n200_v10.000200gold/ag_news_200 gold
ag_keyword_n200_v10.000200gold/ag_news_200 gold
ag_llama_l_n200_v10.000200gold/ag_news_200 gold
ag_llama_m_n200_v10.000200gold/ag_news_200 gold
ag_llama_s_n200_v10.000200gold/ag_news_200 gold
ag_mistral_l_n200_v10.000200gold/ag_news_200 gold
ag_mistral_m_n200_v10.000200gold/ag_news_200 gold
ag_mistral_s_n200_v10.000200gold/ag_news_200 gold
ag_openai_l_n200_v10.000200gold/ag_news_200 gold
ag_openai_m_n200_v10.000200gold/ag_news_200 gold
ag_openai_s_n200_v10.000200gold/ag_news_200 gold
ag_qwen_l_n200_v10.000200gold/ag_news_200 gold
ag_qwen_m_n200_v10.000200gold/ag_news_200 gold
ag_qwen_s_n200_v10.000200gold/ag_news_200 gold
ag_bert_base_fy_n200_v1 · linux0.000200gold/ag_news_200 gold
ag_bert_base_ta_n200_v1 · linux0.000200gold/ag_news_200 gold
ag_bert_mini_n200_v1 · linux0.000200gold/ag_news_200 gold
ag_glm_l_n200_v1 · adjudicated0.090200gold/ag_news_200_adj1 gold
ag_anthropic_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_anthropic_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_anthropic_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_bart_mnli_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_bert_base_fy_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_bert_base_ta_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_bert_mini_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_constant_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_deepseek_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_deepseek_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_deepseek_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_gemma_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_gemma_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_gemma_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_glm_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_glm_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_keyword_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_llama_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_llama_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_llama_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_mistral_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_mistral_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_mistral_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_openai_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_openai_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_openai_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_qwen_l_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_qwen_m_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold
ag_qwen_s_n200_v1 · adjudicated0.000200gold/ag_news_200_adj1 gold

03on dbpedia_280

gold/dbpedia_280db_glm_l_n200_v1: 0.1db_anthropic_s_n200_v1: 0.0071db_deepseek_m_n200_v1: 0.0036db_llama_s_n200_v1: 0.0036db_anthropic_l_n200_v1: 0db_anthropic_m_n200_v1: 0db_bart_mnli_n200_v1: 0db_bert_base_fy_n200_v1: 0db_constant_n200_v1: 0db_deepseek_l_n200_v1: 0db_deepseek_s_n200_v1: 0db_gemma_l_n200_v1: 0db_gemma_m_n200_v1: 0db_gemma_s_n200_v1: 0db_glm_m_n200_v1: 0db_glm_s_n200_v1: 0db_keyword_n200_v1: 0db_llama_l_n200_v1: 0db_llama_m_n200_v1: 0db_mistral_l_n200_v1: 0db_mistral_m_n200_v1: 0db_mistral_s_n200_v1: 0db_openai_l_n200_v1: 0db_openai_m_n200_v1: 0db_openai_s_n200_v1: 0db_qwen_l_n200_v1: 0db_qwen_m_n200_v1: 0db_qwen_s_n200_v1: 0db_bert_base_fy_n200_v1 · linux: 00better · 0.1

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