eval-harness

metric

type_penalty

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
few_nerd_280

01what it measures, and how

untyped_f1 minus f1: the credit lost only to mislabelling entities the arm did find. Large means a typing problem, not a detection problem.

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

02on few_nerd_280

gold/few_nerd_280fn_capitalized_n200_v1: 0.428fn_anthropic_s_n200_v1: 0.144fn_glm_s_n200_v1: 0.133fn_gemma_s_n200_v1: 0.126fn_deepseek_s_n200_v1: 0.124fn_glm_m_n200_v1: 0.121fn_gemma_l_n200_v1: 0.121fn_mistral_s_n200_v1: 0.12fn_llama_l_n200_v1: 0.116fn_llama_s_n200_v1: 0.113fn_deepseek_m_n200_v1: 0.112fn_deepseek_l_n200_v1: 0.107fn_qwen_l_n200_v1: 0.104fn_llama_m_n200_v1: 0.103fn_anthropic_l_n200_v1: 0.102fn_gemma_m_n200_v1: 0.102fn_anthropic_m_n200_v1: 0.101fn_qwen_s_n200_v1: 0.1fn_gliner_n200_v1: 0.098fn_qwen_m_n200_v1: 0.098fn_openai_l_n200_v1: 0.098fn_mistral_l_n200_v1: 0.097fn_mistral_m_n200_v1: 0.096fn_openai_m_n200_v1: 0.092fn_openai_s_n200_v1: 0.091fn_glm_l_n200_v1: 0.087fn_span_marker_n200_v1: 0.07fn_nothing_n200_v1: 00better · 0.428

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