DeepSeek-R1 · AlpacaEval 2.0(DeepSeek-R1 paper Table 3; AlpacaEval 2.0 (LC-winrate); unit not printed) 87.6 · 附件证据(deepseek-r1-ev-002)

DeepSeek-R1 的 AlpacaEval 2.0 · DeepSeek-R1 paper Table 3; AlpacaEval 2.0 (LC-winrate); unit not printed 官方出处。DeepSeek-R1 reported reasoning configuration · 87.6。见《DeepSeek-R1 paper》PDF p.9; Table 3; full header and target row。证据 deepseek-r1-ev-002。

原文与定位

DeepSeek-R1 · AlpacaEval 2.0(DeepSeek-R1 paper Table 3; AlpacaEval 2.0 (LC-winrate); unit not printed),DeepSeek-R1 reported reasoning configuration配置,LC win rate 87.6(具体成绩公布时间未确认)。见《DeepSeek-R1 paper》PDF p.9; Table 3; full header and target row。 查看成绩条件

Benchmark (Metric) | R1-Zero | R1-Dev1 | R1-Dev2 | R1-Dev3 | R1 | AlpacaEval2.0 (LC-winrate) | 24.7 | 50.1 | 55.8 | 62.1 | 87.6 |
官方材料
DeepSeek-R1 paper · PDF 第 9 页
定位
附件 · PDF p.9; Table 3; full header and target row
材料版本
Official DeepSeek-R1 paper: 86/86 PDF pages read; Table 3, Table 8, Table 13, Table 14 and Table 15 checked. Safety, jailbreak and attack sections pp.44-54 were retained as local audit only and excluded from regular mappings.
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本

DeepSeek-R1 · AlpacaEval 2.0(DeepSeek-R1 paper Table 3; AlpacaEval 2.0 (LC-winrate); unit not printed),DeepSeek-R1 reported reasoning configuration配置,LC win rate 87.6(具体成绩公布时间未确认)。官方出处 DeepSeek-R1 paper(附件,Official DeepSeek-R1 paper: 86/86 PDF pages read; Table 3, Table 8, Table 13, Table 14 and Table 15 checked. Safety, jailbreak and attack sections pp.44-54 were retained as local audit only and excluded from regular mappings.)PDF p.9; Table 3; full header and target row,https://arxiv.org/pdf/2501.12948#page=9。Benchlog https://benchlog.lekshome.cc/evidence/r1/deepseek-r1-ev-002/

关联成绩与条件

DeepSeek-R1 · AlpacaEval 2.0 成绩与配置
指标与成绩配置与评测条件证据关系
LC win rate87.6原文 87.6 · 单位 未标注
公布 未确认
DeepSeek-R1 reported reasoning configuration
设置
Maximum generation length 32,768 tokens. For sampled benchmarks: temperature=0.6, top_p=0.95, 64 responses per query to estimate Pass@1.
未知项
Per-benchmark sampling use is not itemized beyond the card preface.
工具
工具条件未记录
推理预算
输出上限 32768 tokens
框架
未记录
协议
Official DeepSeek-R1 paper Table 3; stage table evaluation. Row metric and configuration retained.
条件
DeepSeek-R1 final column; table does not print a percent sign.

直接支持该成绩

在排名中查看此成绩