{"id":1680,"date":"2026-06-04T13:02:04","date_gmt":"2026-06-04T05:02:04","guid":{"rendered":"https:\/\/www.izhuhn.cn\/?p=1680"},"modified":"2026-06-04T15:09:42","modified_gmt":"2026-06-04T07:09:42","slug":"dapo-math-17k-%e6%95%b0%e6%8d%ae%e9%9b%86%e8%b8%a9%e5%9d%91%e8%ae%b0","status":"publish","type":"post","link":"https:\/\/www.izhuhn.cn\/index.php\/2026\/06\/04\/dapo-math-17k-%e6%95%b0%e6%8d%ae%e9%9b%86%e8%b8%a9%e5%9d%91%e8%ae%b0\/","title":{"rendered":"DAPO-Math-17k \u6570\u636e\u96c6\u8e29\u5751\u8bb0"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">\u6700\u8fd1\u5728 GRPO \u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u4e2d\u7528\u5230\u4e86 DAPO-Math-17k \u6570\u636e\u96c6\uff0c\u4e00\u5f00\u59cb\u7406\u6240\u5f53\u7136\u5730 <code>load_dataset(\"BytedTsinghua-SIA\/DAPO-Math-17k\")<\/code>\uff0c\u7ed3\u679c\u8e29\u4e86\u4e24\u4e2a\u5927\u5751\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e00\u3001\u6570\u636e\u91cf\u5f02\u5e38\uff1a1.7 \u4e07\u53d8 170 \u4e07<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">HuggingFace \u4e0a <code>BytedTsinghua-SIA\/DAPO-Math-17k<\/code> \u663e\u793a\u7ea6 1.7M \u884c\uff0c\u540d\u5b57\u91cc\u7684\"17k\"\u8ba9\u4eba\u56f0\u60d1\u3002\u5b9e\u9645\u8dd1\u8d77\u6765\u53d1\u73b0\u4e00\u4e2a Epoch \u8981\u8dd1\u51e0\u5341\u4e2a\u5c0f\u65f6\uff0c\u663e\u5b58\u548c\u65f6\u95f4\u66b4\u6da8\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u539f\u56e0\u5f88\u7b80\u5355\uff1a<strong>\u5b98\u65b9\u4ed3\u5e93\u4e0a\u4f20\u4e86 100 \u500d\u7684\u91cd\u590d\u6570\u636e<\/strong>\u300217,000 \u6761\u539f\u59cb\u6570\u636e\u88ab\u590d\u5236\u7c98\u8d34\u4e86\u7ea6 100 \u6b21\uff0c\u53d8\u6210\u4e86 1,700,000 \u884c\u3002\u5bfc\u81f4\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u663e\u5b58\u4e0e\u65f6\u95f4\u66b4\u6da8<\/strong>\uff1a\u539f\u672c\u51e0\u5c0f\u65f6\u8dd1\u5b8c\u7684 Epoch\uff0c\u5e73\u767d\u65e0\u6545\u591a\u8dd1 100 \u500d<\/li>\n<li><strong>\u6a21\u578b\u4e25\u91cd\u8fc7\u62df\u5408<\/strong>\uff1a\u540c\u4e00\u4e2a Epoch \u5185\u53cd\u590d\u6b7b\u8bb0\u786c\u80cc\u5b8c\u5168\u76f8\u540c\u7684\u6570\u5b66\u9898\uff0c\u6cdb\u5316\u80fd\u529b\u5f7b\u5e95\u5d29\u584c<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u8fd9\u4e5f\u662f\u4e3a\u4ec0\u4e48 Open-R1 \u56e2\u961f\u63a8\u51fa\u4e86 <strong>open-r1\/DAPO-Math-17k-Processed<\/strong>\uff0c\u53bb\u91cd\u540e\u53ea\u4fdd\u7559\u6700\u539f\u59cb\u7684 1.7 \u4e07\u591a\u6761\u552f\u4e00\u6570\u636e\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e8c\u3001\u5b57\u6bb5\u6539\u540d\uff1aprompt \u53d8 source_prompt<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u672c\u4ee5\u4e3a\u5207\u5230 Processed \u7248\u5c31\u4e07\u4e8b\u5927\u5409\uff0c\u7ed3\u679c\u53c8\u62a5 <code>ValueError: num_samples should be a positive integer<\/code>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u539f\u56e0\u662f Processed \u7248\u628a\u5b57\u6bb5\u540d\u4ece <code>prompt<\/code>\uff08\u539f\u7248\uff09\u6539\u6210\u4e86 <code>source_prompt<\/code>\uff0canswer \u5b57\u6bb5\u4e5f\u53d8\u6210\u4e86 <code>solution<\/code>\u3002\u4ee3\u7801\u91cc\u8fd8\u5728\u7528 <code>dataset[\"train\"][\"prompt\"]<\/code>\uff0c\u81ea\u7136\u62ff\u5230\u7a7a\u5217\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e09\u3001\u6b63\u786e\u7528\u6cd5<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u6211\u4f7f\u7528\u7684\u662f <strong>verl<\/strong>\uff08Volcano Engine Reinforcement Learning\uff09\u6846\u67b6\uff0c\u8fd9\u662f\u5b57\u8282\u8df3\u52a8\u5f00\u6e90\u7684 LLM \u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u6846\u67b6\uff0c\u652f\u6301 PPO\/GRPO \u7b49\u7b97\u6cd5\uff0c\u5e95\u5c42\u517c\u5bb9 vLLM \u63a8\u7406\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u65b9\u6848 A\uff1a\u76f4\u63a5\u6539 verl \u6846\u67b6\u53c2\u6570\uff08\u63a8\u8350\uff09<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">verl \u6846\u67b6\u672c\u8eab\u5c31\u652f\u6301\u901a\u8fc7\u914d\u7f6e\u53c2\u6570\u6620\u5c04\u6570\u636e\u5b57\u6bb5\uff0c\u4e0d\u9700\u8981\u6539\u6570\u636e\u96c6\u4e5f\u4e0d\u9700\u8981\u6539\u4ee3\u7801\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># verl \u8bad\u7ec3\u914d\u7f6e\u4e2d\u76f4\u63a5\u6307\u5b9a\u5b57\u6bb5\u6620\u5c04\ndata.prompt_key=source_prompt\ndata.response_key=solution<\/code><\/pre>\n\n\n\n<p class=\"wp-block-paragraph\">\u8fd9\u6837 verl \u4f1a\u81ea\u52a8\u4ece <code>source_prompt<\/code> \u5217\u53d6 prompt\uff0c\u4ece <code>solution<\/code> \u5217\u53d6 response\uff0c\u4e00\u884c\u914d\u7f6e\u641e\u5b9a\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u65b9\u6848 B\uff1aload \u65f6 rename_column\uff08\u5907\u9009\uff09<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c\u4e0d\u60f3\u6539\u6846\u67b6\u914d\u7f6e\uff0c\u4e5f\u53ef\u4ee5\u5728\u52a0\u8f7d\u6570\u636e\u65f6\u505a\u5b57\u6bb5\u91cd\u547d\u540d\uff1a<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>from datasets import load_dataset\n\ndataset = load_dataset(\"open-r1\/DAPO-Math-17k-Processed\")\ndataset = dataset.rename_column(\"source_prompt\", \"prompt\")\ndataset = dataset.rename_column(\"solution\", \"response\")\nprompts = dataset[\"train\"][\"prompt\"]<\/code><\/pre>\n\n\n\n<h2 class=\"wp-block-heading\">\u56db\u3001\u4e24\u4e2a\u5751\u5bf9\u7167<\/h2>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead><tr><th>\u5751<\/th><th>\u73b0\u8c61<\/th><th>\u4fee\u590d<\/th><\/tr><\/thead>\n<tbody>\n<tr><td>\u6570\u636e\u91cd\u590d<\/td><td>\u539f\u7248 1.7M \u884c = 17k \u00d7 ~100 \u6b21\u91cd\u590d<\/td><td>\u7528 open-r1\/DAPO-Math-17k-Processed<\/td><\/tr>\n<tr><td>\u5b57\u6bb5\u6539\u540d<\/td><td>Processed \u7248 prompt \u2192 source_prompt<\/td><td>verl: \u8bbe data.prompt_key \/ response_key<\/td><\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\u603b\u7ed3<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5343\u4e07\u4e0d\u8981\u76f4\u63a5\u7528 <code>BytedTsinghua-SIA\/DAPO-Math-17k<\/code> \u7684 100 \u591a\u4e07\u6761\u8bad\u7ec3<\/li>\n<li>\u52a1\u5fc5\u4f7f\u7528\u53bb\u91cd\u540e\u7684 <code>open-r1\/DAPO-Math-17k-Processed<\/code><\/li>\n<li>verl \u6846\u67b6\u7528\u6237\u76f4\u63a5\u914d <code>data.prompt_key=source_prompt<\/code> + <code>data.response_key=solution<\/code>\uff0c\u96f6\u4ee3\u7801\u6539\u52a8<\/li>\n<li>\u5176\u4ed6\u6846\u67b6\u7528\u6237\u53ef\u7528 <code>rename_column<\/code> \u505a\u5b57\u6bb5\u6620\u5c04<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u7b80\u8bb0\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6700\u8fd1\u5728 GRPO \u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u4e2d\u7528\u5230\u4e86 DAPO-Math-17k \u6570\u636e\u96c6\uff0c\u4e00\u5f00\u59cb\u7406\u6240\u5f53\u7136\u5730 load_dataset(&#8220;Byte &#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1680","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/posts\/1680","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/comments?post=1680"}],"version-history":[{"count":2,"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/posts\/1680\/revisions"}],"predecessor-version":[{"id":1682,"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/posts\/1680\/revisions\/1682"}],"wp:attachment":[{"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/media?parent=1680"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/categories?post=1680"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.izhuhn.cn\/index.php\/wp-json\/wp\/v2\/tags?post=1680"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}