测评结果

[Elisewyh]

我使用您的脚本[tests/test_model_spearman.py]在p40下测试，没有复现出来评测结果，您能帮忙看看怎么回事吗？

model_name encode_type STS-B ATEC BQ LCQMC PAWSX SOHU-dd SOHU-dc avg text2vec-base-chinese FIRST_LAST_AVG 0.921179368 0.2 0.298807152 0.866025404 0.115470054 0.69680998 0.438637106 0.505275581 text2vec-base-chinese LAST_AVG 0.921179368 0.2 0.418330013 0.866025404 0.057735027 0.698363658 0.461319724 0.517564742 text2vec-base-chinese CLS 0.837435789 -0.1 0.298807152 0.866025404 0.230940108 0.672736278 0.435530437 0.463067881 text2vec-base-chinese POOLER 0.865350316 0.3 -0.179284291 0.692820323 0.057735027 0.495812678 0.194071247 0.346643614 text2vec-base-chinese MEAN 0.893264842 0.2 0.358568583 0.866025404 0.173205081 0.700457487 0.447823608 0.519906429 text2vec-base-chinese-paraphrase FIRST_LAST_AVG 0.879307579 0.5 0.836660027 0.866025404 0.577350269 0.771153503 0.621571281 0.721724009 text2vec-base-chinese-paraphrase LAST_AVG 0.879307579 0.5 0.836660027 0.866025404 0.577350269 0.767288163 0.637887383 0.723502689 text2vec-base-chinese-paraphrase CLS 0.688558316 0.4 0.478091444 0.577350269 0.461880215 0.701727277 0.546110893 0.550531202 text2vec-base-chinese-paraphrase POOLER 0.446632421 0.4 0.298807152 0.519615242 0.346410162 0.563723704 0.393554815 0.424106214 text2vec-base-chinese-paraphrase MEAN 0.879307579 0.5 0.836660027 0.866025404 0.75055535 0.76531092 0.639699763 0.74822272 text2vec-base-chinese-sentence FIRST_LAST_AVG 0.865350316 0.5 0.717137166 0.866025404 0.577350269 0.694734627 0.460195945 0.668684818 text2vec-base-chinese-sentence LAST_AVG 0.879307579 0.5 0.717137166 0.866025404 0.692820323 0.701333547 0.530489951 0.698159138 text2vec-base-chinese-sentence CLS 0.781606737 0.5 0.776898596 0.808290377 0.519615242 0.512156235 0.198754203 0.585331627 text2vec-base-chinese-sentence POOLER 0.795564 0.5 0.657375735 0.866025404 0.346410162 0.375155273 0.082185259 0.517530833 text2vec-base-chinese-sentence MEAN 0.851393053 0.5 0.776898596 0.866025404 0.519615242 0.705212815 0.530814109 0.678565603

[shibing624]

没看到问题重点是啥？差多少，哪个方法差？

[Elisewyh]

没看到问题重点是啥？差多少，哪个方法差？

就是跑出来的结果和你这个测评报告里的结果差异很大 https://huggingface.co/shibing624/text2vec-base-chinese