eval-harness

metric

answer_words

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

How many whitespace-separated words the reply had: a description, not a quality.

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

02on ag_news_200

gold/ag_news_200ag_glm_l_n200_v1: 6.045ag_anthropic_l_n200_v1: 1ag_anthropic_m_n200_v1: 1ag_anthropic_s_n200_v1: 1ag_bart_mnli_n200_v1: 1ag_bert_base_fy_n200_v1: 1ag_bert_base_ta_n200_v1: 1ag_bert_mini_n200_v1: 1ag_constant_n200_v1: 1ag_deepseek_l_n200_v1: 1ag_deepseek_m_n200_v1: 1ag_deepseek_s_n200_v1: 1ag_gemma_l_n200_v1: 1ag_gemma_m_n200_v1: 1ag_gemma_s_n200_v1: 1ag_glm_m_n200_v1: 1ag_glm_s_n200_v1: 1ag_keyword_n200_v1: 1ag_llama_l_n200_v1: 1ag_llama_m_n200_v1: 1ag_llama_s_n200_v1: 1ag_mistral_l_n200_v1: 1ag_mistral_m_n200_v1: 1ag_mistral_s_n200_v1: 1ag_openai_l_n200_v1: 1ag_openai_m_n200_v1: 1ag_openai_s_n200_v1: 1ag_qwen_l_n200_v1: 1ag_qwen_m_n200_v1: 1ag_qwen_s_n200_v1: 1ag_bert_base_fy_n200_v1 · linux: 1ag_bert_base_ta_n200_v1 · linux: 1ag_bert_mini_n200_v1 · linux: 1gold/ag_news_200_adj1ag_glm_l_n200_v1 · adjudicated: 6.045ag_anthropic_l_n200_v1 · adjudicated: 1ag_anthropic_m_n200_v1 · adjudicated: 1ag_anthropic_s_n200_v1 · adjudicated: 1ag_bart_mnli_n200_v1 · adjudicated: 1ag_bert_base_fy_n200_v1 · adjudicated: 1ag_bert_base_ta_n200_v1 · adjudicated: 1ag_bert_mini_n200_v1 · adjudicated: 1ag_constant_n200_v1 · adjudicated: 1ag_deepseek_l_n200_v1 · adjudicated: 1ag_deepseek_m_n200_v1 · adjudicated: 1ag_deepseek_s_n200_v1 · adjudicated: 1ag_gemma_l_n200_v1 · adjudicated: 1ag_gemma_m_n200_v1 · adjudicated: 1ag_gemma_s_n200_v1 · adjudicated: 1ag_glm_m_n200_v1 · adjudicated: 1ag_glm_s_n200_v1 · adjudicated: 1ag_keyword_n200_v1 · adjudicated: 1ag_llama_l_n200_v1 · adjudicated: 1ag_llama_m_n200_v1 · adjudicated: 1ag_llama_s_n200_v1 · adjudicated: 1ag_mistral_l_n200_v1 · adjudicated: 1ag_mistral_m_n200_v1 · adjudicated: 1ag_mistral_s_n200_v1 · adjudicated: 1ag_openai_l_n200_v1 · adjudicated: 1ag_openai_m_n200_v1 · adjudicated: 1ag_openai_s_n200_v1 · adjudicated: 1ag_qwen_l_n200_v1 · adjudicated: 1ag_qwen_m_n200_v1 · adjudicated: 1ag_qwen_s_n200_v1 · adjudicated: 11better · 6.045

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 answer_wordsitemsscored against
ag_glm_l_n200_v16.045200gold/ag_news_200 gold
ag_anthropic_l_n200_v11.000200gold/ag_news_200 gold
ag_anthropic_m_n200_v11.000200gold/ag_news_200 gold
ag_anthropic_s_n200_v11.000200gold/ag_news_200 gold
ag_bart_mnli_n200_v11.000200gold/ag_news_200 gold
ag_bert_base_fy_n200_v11.000200gold/ag_news_200 gold
ag_bert_base_ta_n200_v11.000200gold/ag_news_200 gold
ag_bert_mini_n200_v11.000200gold/ag_news_200 gold
ag_constant_n200_v11.000200gold/ag_news_200 gold
ag_deepseek_l_n200_v11.000200gold/ag_news_200 gold
ag_deepseek_m_n200_v11.000200gold/ag_news_200 gold
ag_deepseek_s_n200_v11.000200gold/ag_news_200 gold
ag_gemma_l_n200_v11.000200gold/ag_news_200 gold
ag_gemma_m_n200_v11.000200gold/ag_news_200 gold
ag_gemma_s_n200_v11.000200gold/ag_news_200 gold
ag_glm_m_n200_v11.000200gold/ag_news_200 gold
ag_glm_s_n200_v11.000200gold/ag_news_200 gold
ag_keyword_n200_v11.000200gold/ag_news_200 gold
ag_llama_l_n200_v11.000200gold/ag_news_200 gold
ag_llama_m_n200_v11.000200gold/ag_news_200 gold
ag_llama_s_n200_v11.000200gold/ag_news_200 gold
ag_mistral_l_n200_v11.000200gold/ag_news_200 gold
ag_mistral_m_n200_v11.000200gold/ag_news_200 gold
ag_mistral_s_n200_v11.000200gold/ag_news_200 gold
ag_openai_l_n200_v11.000200gold/ag_news_200 gold
ag_openai_m_n200_v11.000200gold/ag_news_200 gold
ag_openai_s_n200_v11.000200gold/ag_news_200 gold
ag_qwen_l_n200_v11.000200gold/ag_news_200 gold
ag_qwen_m_n200_v11.000200gold/ag_news_200 gold
ag_qwen_s_n200_v11.000200gold/ag_news_200 gold
ag_bert_base_fy_n200_v1 · linux1.000200gold/ag_news_200 gold
ag_bert_base_ta_n200_v1 · linux1.000200gold/ag_news_200 gold
ag_bert_mini_n200_v1 · linux1.000200gold/ag_news_200 gold
ag_glm_l_n200_v1 · adjudicated6.045200gold/ag_news_200_adj1 gold
ag_anthropic_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_anthropic_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_anthropic_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_bart_mnli_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_bert_base_fy_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_bert_base_ta_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_bert_mini_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_constant_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_deepseek_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_deepseek_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_deepseek_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_gemma_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_gemma_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_gemma_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_glm_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_glm_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_keyword_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_llama_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_llama_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_llama_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_mistral_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_mistral_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_mistral_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_openai_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_openai_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_openai_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_qwen_l_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_qwen_m_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold
ag_qwen_s_n200_v1 · adjudicated1.000200gold/ag_news_200_adj1 gold

03on dbpedia_280

gold/dbpedia_280db_glm_l_n200_v1: 5.679db_anthropic_s_n200_v1: 1.211db_llama_s_n200_v1: 1.111db_deepseek_m_n200_v1: 1.068db_gemma_s_n200_v1: 1.043db_anthropic_l_n200_v1: 1db_anthropic_m_n200_v1: 1db_bart_mnli_n200_v1: 1db_bert_base_fy_n200_v1: 1db_constant_n200_v1: 1db_deepseek_l_n200_v1: 1db_deepseek_s_n200_v1: 1db_gemma_l_n200_v1: 1db_gemma_m_n200_v1: 1db_glm_m_n200_v1: 1db_keyword_n200_v1: 1db_llama_l_n200_v1: 1db_llama_m_n200_v1: 1db_mistral_l_n200_v1: 1db_mistral_m_n200_v1: 1db_mistral_s_n200_v1: 1db_openai_l_n200_v1: 1db_openai_m_n200_v1: 1db_openai_s_n200_v1: 1db_qwen_l_n200_v1: 1db_qwen_m_n200_v1: 1db_qwen_s_n200_v1: 1db_bert_base_fy_n200_v1 · linux: 1db_glm_s_n200_v1: 0.9960.996better · 5.679

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