DeepSeek R1 Distill Qwen · GPQA Diamond · Diamond subset; task snapshot unspecified; table does not print a percent sign · 历代记录

按发布时间查看 DeepSeek R1 Distill Qwen 的 4 次发布在 GPQA Diamond · Diamond subset; task snapshot unspecified; table does not print a percent sign 的成绩、配置与披露。

仅为直接前代间的唯一、有效精确值计算差值,要求单位、指标与明确记录的配置和条件一致。

DeepSeek-R1-Distill-Qwen-7B

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

DeepSeek-R1-Distill-Qwen-7B · GPQA Diamond 成绩与配置
指标与成绩配置与评测条件出处
Pass@149.1原文 49.1 · 单位 未标注
公布 未确认
DeepSeek-R1-Distill-Qwen-7B
设置
Distilled dense checkpoint; source card preface states max generation length 32,768 tokens, temperature=0.6, top_p=0.95, and 64 responses per query for sampled pass@1 estimates.
未知项
The source does not specify a separate effort value for each distilled checkpoint.
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
未说明
条件
Official distilled-model evaluation table; exact model scale retained in the release name and source row. The source table does not print a percent sign.
在排名中查看此成绩

DeepSeek-R1-Distill-Qwen-32B

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

DeepSeek-R1-Distill-Qwen-32B · GPQA Diamond 成绩与配置
指标与成绩配置与评测条件出处
Pass@162.1原文 62.1 · 单位 未标注
公布 未确认
DeepSeek-R1-Distill-Qwen-32B
设置
Distilled dense checkpoint; source card preface states max generation length 32,768 tokens, temperature=0.6, top_p=0.95, and 64 responses per query for sampled pass@1 estimates.
未知项
The source does not specify a separate effort value for each distilled checkpoint.
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
未说明
条件
Official distilled-model evaluation table; exact model scale retained in the release name and source row. The source table does not print a percent sign.
在排名中查看此成绩

DeepSeek-R1-Distill-Qwen-14B

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

DeepSeek-R1-Distill-Qwen-14B · GPQA Diamond 成绩与配置
指标与成绩配置与评测条件出处
Pass@159.1原文 59.1 · 单位 未标注
公布 未确认
DeepSeek-R1-Distill-Qwen-14B
设置
Distilled dense checkpoint; source card preface states max generation length 32,768 tokens, temperature=0.6, top_p=0.95, and 64 responses per query for sampled pass@1 estimates.
未知项
The source does not specify a separate effort value for each distilled checkpoint.
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
未说明
条件
Official distilled-model evaluation table; exact model scale retained in the release name and source row. The source table does not print a percent sign.
在排名中查看此成绩

DeepSeek-R1-Distill-Qwen-1.5B

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

DeepSeek-R1-Distill-Qwen-1.5B · GPQA Diamond 成绩与配置
指标与成绩配置与评测条件出处
Pass@133.8原文 33.8 · 单位 未标注
公布 未确认
DeepSeek-R1-Distill-Qwen-1.5B
设置
Distilled dense checkpoint; source card preface states max generation length 32,768 tokens, temperature=0.6, top_p=0.95, and 64 responses per query for sampled pass@1 estimates.
未知项
The source does not specify a separate effort value for each distilled checkpoint.
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
未说明
条件
Official distilled-model evaluation table; exact model scale retained in the release name and source row. The source table does not print a percent sign.
在排名中查看此成绩