{"id":812618,"date":"2026-07-16T16:32:57","date_gmt":"2026-07-16T08:32:57","guid":{"rendered":"https:\/\/www.alphaseek.cn\/?p=812618"},"modified":"2026-07-16T16:32:57","modified_gmt":"2026-07-16T08:32:57","slug":"rlinf-v0-3%e6%ad%a3%e5%bc%8f%e5%8f%91%e5%b8%83%ef%bc%8c%e4%b8%80%e7%ab%99%e5%bc%8f%e5%85%b7%e8%ba%ab%e6%99%ba%e8%83%bd%e5%bc%80%e5%8f%91%e5%b9%b3%e5%8f%b0%e6%9d%a5%e4%ba%86","status":"publish","type":"post","link":"https:\/\/www.alphaseek.cn\/?p=812618","title":{"rendered":"RLinf v0.3\u6b63\u5f0f\u53d1\u5e03\uff0c\u4e00\u7ad9\u5f0f\u5177\u8eab\u667a\u80fd\u5f00\u53d1\u5e73\u53f0\u6765\u4e86"},"content":{"rendered":"<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/5DB16DF3442ACD9CAEF2E306A8C338AE4ED875F9_size51_w900_h383.jpg\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"RLinf v0.3\u6b63\u5f0f\u53d1\u5e03\uff0c\u4e00\u7ad9\u5f0f\u5177\u8eab\u667a\u80fd\u5f00\u53d1\u5e73\u53f0\u6765\u4e86\"  \/><\/p>\n<p><strong>\u673a\u5668\u4eba\u524d\u77bb\uff08\u516c\u4f17\u53f7\uff1arobot_pro\uff09<\/strong><\/p>\n<p><strong>\u4f5c\u8005 | \u5468\u52a0\u7426<\/strong><\/p>\n<p><strong>\u7f16\u8f91 | \u6f20\u5f71<\/strong><\/p>\n<p>\u673a\u5668\u4eba\u524d\u77bb7\u670816\u65e5\u62a5\u9053\uff0c\u4eca\u5929\uff0c\u5927\u89c4\u6a21\u5f3a\u5316\u5b66\u4e60\u57fa\u7840\u8bbe\u65bd\u9879\u76eeRLinf\uff0c<strong>\u6b63\u5f0f\u5347\u7ea7\u4e3aRLinf v0.3<\/strong>\u3002RLinf\u662f\u7531<strong>\u65e0\u95ee\u82af\u7a79\u8054\u5408\u6e05\u534e\u5927\u5b66\u7b49<\/strong>\u5171\u540c\u7814\u53d1\u3001\u5168\u7403\u9996\u4e2a\u9762\u5411\u5177\u8eab\u667a\u80fd\u6301\u7eed\u8fdb\u5316\u7684\u9879\u76ee\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/7C0DF26909BFE1A9B638C475500860EFC14949AF_size101_w640_h142.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"RLinf v0.3\u6b63\u5f0f\u53d1\u5e03\uff0c\u4e00\u7ad9\u5f0f\u5177\u8eab\u667a\u80fd\u5f00\u53d1\u5e73\u53f0\u6765\u4e86\" \/><\/p>\n<p>\u25b2\uff08\u56fe\u6e90\uff1a\u65e0\u95ee\u82af\u7a79\uff09<\/p>\n<p>\u6b64\u524d\u5df2\u53d1\u5e03\u7684RLinf v0.1\u548cRLinf v0.2\uff0c\u5206\u522b\u5b8c\u6210\u4e86\u5f3a\u5316\u5b66\u4e60\u7cfb\u7edf\u62bd\u8c61\u4e0e\u771f\u5b9e\u4e16\u754c\u5728\u7ebf\u5b66\u4e60\u57fa\u7840\u8bbe\u65bd\u5efa\u8bbe\u3002<\/p>\n<p>RLinf v0.3\u5219\u5347\u7ea7\u4e3a\u9762\u5411\u5177\u8eab\u667a\u80fd\u6301\u7eed\u8fdb\u5316\u7684<strong>\u4e00\u7ad9\u5f0f\u5f00\u53d1\u5e73\u53f0<\/strong>\uff0c\u6253\u901a\u6570\u636e\u91c7\u96c6\u3001\u6570\u636e\u7ba1\u7406\u3001\u76d1\u7763\u5fae\u8c03\uff08SFT\uff09\u3001\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u3001\u6a21\u578b\u8bc4\u6d4b\u4ee5\u53ca\u771f\u673a\u90e8\u7f72\u7b49\u73af\u8282\uff0c\u5b9e\u73b0\u4ece\u4eff\u771f\u8bad\u7ec3\u5230\u771f\u5b9e\u673a\u5668\u4eba\u5728\u7ebf\u5b66\u4e60\u3001\u518d\u5230\u6301\u7eed\u8fed\u4ee3\u4f18\u5316\u7684\u7edf\u4e00\u5f00\u53d1\u95ed\u73af\u3002<\/p>\n<p>Github\u94fe\u63a5\uff1a<\/p>\n<p>https:\/\/github.com\/RLinf\/RLinf<\/p>\n<p>RLinf Tech Report\uff1a<\/p>\n<p>https:\/\/arxiv.org\/abs\/2509.15965<\/p>\n<p>\u4e00\u3001\u4e94\u5927\u7ef4\u5ea6\u5347\u7ea7\uff0c\u4ece\u6a21\u578b\u5230\u771f\u673a\u90e8\u7f72\u4e00\u7ad9\u5f0f\u89e3\u51b3<\/p>\n<p>\u6b64\u6b21\u53d1\u5e03\u7684RLinf v0.3\uff0c\u4ece\u6a21\u578b\u3001\u7b97\u6cd5\u3001\u771f\u673a\u3001\u4eff\u771f\u3001\u7cfb\u7edf\u4e94\u4e2a\u7ef4\u5ea6\u8fdb\u884c\u5347\u7ea7\uff0c\u4e3a\u5f00\u53d1\u8005\u964d\u4f4e\u4e86\u5177\u8eab\u667a\u80fd\u5f00\u53d1\u95e8\u69db\u3002<\/p>\n<p><strong>\u5728\u6a21\u578b\u65b9\u9762<\/strong>\uff0c\u65b0\u589e<strong>6\u6b3e\u5177\u8eab\u6a21\u578b<\/strong>\uff0c\u6db5\u76d6\u4e16\u754c\u6a21\u578b\u3001VLA\u6a21\u578b\u548c\u7cfb\u7edf\u7ea7\u52a0\u901f\u3002<\/p>\n<p>\u5305\u542bDexbotic DM0\u6a21\u578b\u3001DreamZero\u6a21\u578b\u3001GR00T-N1.6\/N1.7\u6a21\u578bRL\u5fae\u8c03\u3001ABot-M0\u6a21\u578b\u3001StarVLA\u6a21\u578b\uff08GRPO on LIBERO\uff09\u3001LingBot-VLA\u6a21\u578b\uff08RoboTwin\u73af\u5883SFT\/RL\uff09\u3002<\/p>\n<p>\u5176\u4e2dDreamZero\u6a21\u578b\uff0c\u57fa\u4e8eWAN2.1\/2.2\u89c6\u9891\u751f\u6210\u4e16\u754c\u6a21\u578b\u5fae\u8c03\u7684VLA<a href=\"https:\/\/www.alphaseek.cn\/?tag=%e7%ad%96%e7%95%a5\" title=\"View all posts in \u7b56\u7565\" target=\"_blank\" class=\"tag_link\">\u7b56\u7565<\/a>\uff0c\u96c6\u6210\u8fdbSFT\u5de5\u4f5c\u6d41\uff0c\u901a\u8fc7FSDP2\/CUDA Graph\u7b49\u7cfb\u7edf\u7ea7\u52a0\u901f\u53d6\u5f97\u8fd14\u500d\u541e\u5410\u63d0\u5347\u3002<\/p>\n<\/p>\n<p>\u25b2DreamZero\uff08\u56fe\u6e90\uff1aRLinf\uff09<\/p>\n<p><strong>\u5728\u7b97\u6cd5\u65b9\u9762<\/strong>\uff0cRLinf v0.3\u5728<strong>\u771f\u673aRL<\/strong>\u3001<strong>\u4eff\u771fRL<\/strong>\u548c<strong>\u4eba\u5728\u73af\u5b66\u4e60<\/strong>\u4e09\u4e2a\u65b9\u5411\u6709\u65b0\u589e\u3002<\/p>\n<p>\u771f\u673a\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u5c06DSRL\u6269\u5c55\u81f3Pi0.5\u6a21\u578b\uff0c\u53ef\u5728\u771f\u673a\u4e0a\u9ad8\u6548\u81ea\u6211\u8fed\u4ee3\uff0c\u5927\u5e45\u63d0\u5347\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u8fd8\u65b0\u589eRECAP\uff08\u57fa\u4e8e\u79bb\u7ebf\u4f18\u52bf\u4f30\u8ba1\u7684\u7b56\u7565\u4f18\u5316\uff09\u8bad\u7ec3\u6d41\u6c34\u7ebf\u652f\u6301\uff0cSAC-Flow\u7b97\u6cd5\u652f\u6301\uff0c\u5e76\u6269\u5c55\u5230DOS-W1\u7b49\u771f\u673a\u573a\u666f\uff0c\u8fdb\u4e00\u6b65\u964d\u4f4e\u6a21\u578b\u771f\u673a\u4f18\u5316\u7684\u95e8\u69db\u3002<\/p>\n<p>\u4eff\u771f\u5f3a\u5316\u7b97\u6cd5\u5728RLinf v0.2\u7684\u57fa\u7840\u4e0a\u6269\u5c55\u652f\u6301MLP\u7b49\u65b0\u7b56\u7565\uff0c\u65b0\u589easync DSRL\u914d\u7f6e\uff0c\u8bad\u7ec3\u8c03\u5ea6\u66f4\u7075\u6d3b\uff1b\u65b0\u589eD4RL\u79bb\u7ebfIQL\u8bad\u7ec3\u652f\u6301\uff0c\u53ef\u5728\u4eff\u771f\u73af\u5883\u4e2d\uff0c\u7528\u5b58\u91cf\u79bb\u7ebf\u6570\u636e\u5b8c\u6210\u6a21\u578b\u8bad\u7ec3\u3002<\/p>\n<p>\u4eba\u5728\u73af\u5b66\u4e60\u8986\u76d6\u4eff\u771f\u548c\u771f\u673a\u573a\u666f\uff0c\u65b0\u589eDAgger\u5728\u7ebf\u6a21\u4eff\u5b66\u4e60\u7b97\u6cd5\u652f\u6301\u548cHG-DAgger\u771f\u673a\u5728\u7ebf\u8bad\u7ec3\u652f\u6301\uff0c\u5927\u5e45\u964d\u4f4e\u673a\u5668\u4eba\u5b9e\u64cd\u5b66\u4e60\u96be\u5ea6\u3002<\/p>\n<p><strong>\u5728\u771f\u673a\u65b9\u9762<\/strong>\uff0c\u6253\u901a\u6570\u636e\u91c7\u96c6\u5230\u771f\u673a\u90e8\u7f72\u7684\u95ed\u73af\u3002<\/p>\n<p>\u6570\u636e\u91c7\u96c6\u90e8\u5206\uff0c<strong>\u65b0\u589e3\u79cd\u9065\u64cd\u4f5c\u65b9\u5f0f<\/strong>\uff0c\u5305\u542b\u7a7a\u95f4\u9f20\u6807\u9065\u64cd\u4f5c\u3001VR\u9065\u64cd\u4f5c\u3001GELLO\u9065\u64cd\u4f5c\u3002\u8bad\u7ec3\u94fe\u8def\u90e8\u5206\uff0c\u652f\u6301LeRobot\u683c\u5f0f\u6570\u636e\u91c7\u96c6\uff0c\u4fbf\u4e8e\u4e0eHuggingFace LeRobot\u751f\u6001\u4e92\u901a\uff0c\u540c\u65f6\u8fd8\u652f\u6301Pi0\u771f\u673aSFT\u90e8\u7f72\u3001\u771f\u673areward model\u6570\u636e\u91c7\u96c6\u3002<\/p>\n<p>\u5728\u771f\u673a\u5e73\u53f0\u548c\u672b\u7aef\u6267\u884c\u5668\u4e0a\uff0c<strong>\u65b0\u589e3\u6b3e\u771f\u673a\u5e73\u53f0<\/strong>\uff0c\u5305\u542b\u53cc\u81c2Franka\u5e73\u53f0\u3001GimArm\u548cDOS-W1\u771f\u673a\u5e73\u53f0\uff1b\u65b0\u589e2\u6b3e\u672b\u7aef\u6267\u884c\u5668\uff0c\u5305\u542bFranka DexHand\u7075\u5de7\u624b\u3001Franka Robotiq\u5939\u722a\u540e\u7aef\u3001Franka Robotiq\u53caZED\/LUMOS V4L2\u76f8\u673a\u4e0e\u5939\u722a\u540e\u7aef\u3002<\/p>\n<p>\u6b64\u5916\uff0c<strong>\u5728\u4eff\u771f\u65b9\u9762<\/strong>\uff0cRLinf v0.3\u63d0\u5347\u573a\u666f\u8986\u76d6\uff0c<strong>\u65b0\u589e5\u79cd\u4eff\u771f\u5668<\/strong>\uff0c\u5305\u542bGenesis\u4eff\u771f\u5668\u3001Polaris\u4eff\u771f\u5668\u3001RoboVerse\u4eff\u771f\u5668\u3001Libero+\/LiberoPro\u53d8\u4f53\u73af\u5883\u3001Embodichain\uff08CartPole\uff09\u73af\u5883\u3001IsaacLab\u4e0a\u03c00.5 PPO finetuning\u3001RoboCasa close-drawer\u7b49RL\u793a\u4f8b\uff0c\u8fd8\u5b8c\u5584Behavior\u73af\u5883\u652f\u6301\u3002<\/p>\n<p><strong>\u5728\u7cfb\u7edf\u65b9\u9762<\/strong>\uff0cRLinf v0.3\u65b0\u589eReward Model\u3001Value Model\u3001SGLang\u63a8\u7406\u670d\u52a1\u5316\u548c\u89e3\u8026\u73af\u5883\u6267\u884c\u7ec4\u4ef6\uff0c\u540c\u65f6\u8fd8\u6253\u901a\u6607\u817eAscend\u3001AMD\u7b49\u56fd\u4ea7\u53ca\u5f02\u6784AI\u8ba1\u7b97\u5e73\u53f0\u3002<\/p>\n<p>\u6b64\u5916\uff0cRLinf v0.3\u4e3a\u667a\u80fd\u4f53RL\u573a\u666f\u63d0\u4f9b\u4e86\u66f4\u5f3a\u5927\u7684\u8bad\u7ec3\u548c\u8bc4\u6d4b\u57fa\u7840\uff0c\u65b0\u589eAgentLightning\u591a\u8f6e\u5355\u667a\u80fd\u4f53RL\u8bad\u7ec3\u4e0eCalc-X\u8bc4\u6d4b\u3001Megatron-Bridge actor\u540e\u7aef\u4e0eWideSeek judge\u7684\u5185\u7f6e sglang\u7684\u652f\u6301\u3002<\/p>\n<p>\u7ed3\u8bed\uff1a\u8fdb\u5316\u4e0d\u6b62\u4e8e\u6b64\uff0c\u6301\u7eed\u6253\u78e8\u5b9e\u7528\u5e73\u53f0<\/p>\n<p>RLinf\u81ea\u5f00\u6e90\u4ee5\u6765\uff0c\u5728GitHub\u4e0a\u5df2\u83b7\u5f974100+Stars\u3001600+Forks\u3001100+Contributors\uff0c\u5e76\u4e14\u88abIsaac Lab\u5b98\u65b9\u6536\u5f55\u4e3a\u5176\u9996\u4e2a\u9762\u5411\u5177\u8eab\u5927\u6a21\u578b\u7684\u8bad\u7ec3\u5f15\u64ce\u3002<\/p>\n<p>RLinf\u7ecf\u8fc7\u4e09\u6b21\u7248\u672c\u8fed\u4ee3\uff0c\u8d70\u51fa\u4e00\u6761\u4ece\u201c\u597d\u7528\u201d\u5230\u201c\u6301\u7eed\u8fdb\u5316\u201d\u7684\u6e05\u6670\u8def\u5f84\u3002\u6b64\u6b21\u5347\u7ea7\u4e0d\u4ec5\u4ec5\u662f\u6a21\u578b\u5e93\u7684\u6269\u5145\u6216\u7b97\u6cd5\u7684\u53e0\u52a0\uff0c\u66f4\u662f\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4e86\u4e00\u6761\u201c\u5f00\u7bb1\u5373\u7528\u201d\u65b9\u6848\u3002\u540e\u7eed\u7248\u672c\u7684\u6f14\u8fdb\u65b9\u5411\uff0cRLinf\u56e2\u961f\u5c06\u7ed3\u5408\u5f00\u53d1\u8005\u5728\u5b9e\u9645\u9879\u76ee\u4e2d\u7684\u53cd\u9988\u548c\u9700\u6c42\uff0c\u6301\u7eed\u4f18\u5316\u7cfb\u7edf\u6027\u80fd\u548c\u7b97\u6cd5\u8986\u76d6\u3002<\/p>\n<p>(\u6765\u6e90\uff1a\u65b0\u6d6a\u79d1\u6280)<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u673a\u5668\u4eba\u524d\u77bb\uff08\u516c\u4f17\u53f7\uff1arobot_pro\uff09 \u4f5c\u8005 | \u5468\u52a0\u7426 \u7f16\u8f91 | \u6f20\u5f71 \u673a\u5668\u4eba\u524d\u77bb7\u670816\u65e5\u62a5\u9053\uff0c\u4eca\u5929\uff0c\u5927 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[6],"tags":[88],"views":3,"_links":{"self":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts\/812618"}],"collection":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=812618"}],"version-history":[{"count":0,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts\/812618\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=812618"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=812618"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=812618"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}