
{"id":11553,"date":"2026-08-27T17:48:32","date_gmt":"2026-08-27T09:48:32","guid":{"rendered":"https:\/\/infernews.com\/blog\/streaming-multimodal-temporal-modeling-for-vision-language-action-models\/"},"modified":"2026-08-27T17:56:13","modified_gmt":"2026-08-27T09:56:13","slug":"streaming-multimodal-temporal-modeling-for-vision-language-action-models","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/streaming-multimodal-temporal-modeling-for-vision-language-action-models\/","title":{"rendered":"StreamPI \u6a5f\u68b0\u4eba\u6a21\u578b\u7684\u8a18\u61b6\u7cfb\u7d71"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-47434a582575.jpg\" alt=\"Overview of the StreamPI architecture and streaming inference workflow\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u6a5f\u68b0\u4eba\u57f7\u884c\u6293\u53d6\u3001\u63d2\u5165\u7b49\u9023\u7e8c\u4efb\u52d9\u6642\uff0c\u53ea\u770b\u7576\u524d\u756b\u9762\u5bb9\u6613\u5931\u53bb\u7269\u4ef6\u4f4d\u7f6e\u8b8a\u5316\uff1bStreamPI \u5c07\u6bcf\u7d44\u8996\u89ba\u89c0\u5bdf\u8207\u8a9e\u8a00\u6307\u4ee4\u8996\u70ba\u4e00\u500b\u6642\u5e8f\u55ae\u5143\uff0c\u5c6c\u65bc\u70ba Vision-Language-Action (VLA) \u6a21\u578b\u52a0\u5165\u4e32\u6d41\u6642\u5e8f\u63a8\u7406\u80fd\u529b\u7684\u6846\u67b6\uff0c\u91dd\u5c0d\u7684\u6b63\u662f\u55ae\u5e40 \u03c0\u2080.\u2085 \u7121\u6cd5\u4fdd\u7559\u6b77\u53f2\u89c0\u5bdf\u7684\u9650\u5236\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u6c92\u6709\u984d\u5916\u52a0\u5165\u6a21\u578b\u53c3\u6578\uff0c\u800c\u662f\u628a\u591a\u7d44 token \u63a5\u5728\u540c\u4e00\u5e8f\u5217\uff0c\u518d\u7528 block-wise attention mask \u63a7\u5236\u8cc7\u8a0a\u6d41\u3002\u6bcf\u500b\u55ae\u5143\u5167\u63a1\u7528\u96d9\u5411\u6ce8\u610f\u529b\uff0c\u8b93\u8996\u89ba\u8207\u8a9e\u8a00\u5145\u5206\u878d\u5408\uff1b\u55ae\u5143\u4e4b\u9593\u63a1\u7528\u56e0\u679c\u6ce8\u610f\u529b\uff0c\u914d\u5408 KV cache \u4fdd\u7559\u904e\u53bb\u5167\u5bb9\uff0c\u907f\u514d\u6bcf\u6b21\u91cd\u65b0\u8655\u7406\u5b8c\u6574\u6b77\u53f2\u3002\u91cd\u8907\u653e\u5165\u8a9e\u8a00\u6307\u4ee4\u4ea6\u53ef\u4ee4\u4efb\u52d9\u76ee\u6a19\u5728\u8996\u89ba\u5167\u5bb9\u589e\u52a0\u5f8c\u4fdd\u6301\u6e05\u6670\u3002<\/p>\n\n\n<div\n    \tclass=\"align wp-block-vpb-video\"    id='vpbpVideoPlayer-1'\n    data-attributes='{&quot;source&quot;:&quot;https:\\\/\\\/happinesslz.github.io\\\/projects\\\/StreamPI\\\/StreamPI_files\\\/streampi_pick.mp4&quot;,&quot;repeat&quot;:true,&quot;autoplay&quot;:true,&quot;muted&quot;:true,&quot;align&quot;:&quot;&quot;,&quot;poster&quot;:&quot;&quot;,&quot;controls&quot;:{&quot;play-large&quot;:true,&quot;restart&quot;:false,&quot;rewind&quot;:true,&quot;play&quot;:true,&quot;fast-forward&quot;:true,&quot;progress&quot;:true,&quot;current-time&quot;:true,&quot;duration&quot;:false,&quot;mute&quot;:true,&quot;volume&quot;:true,&quot;pip&quot;:false,&quot;airplay&quot;:false,&quot;captions&quot;:false,&quot;settings&quot;:true,&quot;download&quot;:false,&quot;fullscreen&quot;:true},&quot;captions&quot;:[],&quot;ratio&quot;:&quot;16:9&quot;,&quot;accentColor&quot;:&quot;&quot;,&quot;lazyLoad&quot;:false,&quot;width&quot;:&quot;100%&quot;,&quot;radius&quot;:&quot;0px&quot;,&quot;resetOnEnd&quot;:false,&quot;autoHideControl&quot;:true,&quot;isSetup&quot;:false}'\n><\/div>\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ee5\u9810\u8a13\u7df4\u55ae\u5e40\u6a21\u578b\u5ef6\u4f38\u81f3\u55ae\u5e40\u6216\u591a\u5e40\u63a8\u7406<\/li>\n\n\n\n<li>\u4e0d\u589e\u52a0\u53c3\u6578\uff0c\u652f\u63f4\u5f48\u6027\u4e0a\u4e0b\u6587\u9577\u5ea6<\/li>\n\n\n\n<li>\u4ee5 random-interval sampling \u914d\u5408 temporal masking\uff0c\u6a21\u64ec\u975e\u56fa\u5b9a\u89c0\u5bdf\u6642\u9593<\/li>\n\n\n\n<li>KV cache \u9069\u5408\u4e32\u6d41\u63a8\u7406\u53ca\u975e\u540c\u6b65\u6a5f\u68b0\u4eba\u63a7\u5236<\/li>\n\n\n\n<li>RTX 4090 \u7531\u4e00\u5e40\u589e\u81f3\u4e94\u5e40\uff0c\u5e73\u5747\u5ef6\u9072\u53ea\u7531 94.4 ms \u5347\u81f3 103.6 ms<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u8a13\u7df4\u6642\u52a0\u5165\u96a8\u6a5f\u5e40\u9593\u9694\uff0c\u8b93\u6a21\u578b\u63a5\u89f8\u4e0d\u540c\u89c0\u5bdf\u7bc0\u594f\uff1b\u7814\u7a76\u8cc7\u6599\u4ea6\u63d0\u5230\u6bcf\u4e09\u5e40\u53d6\u6a23\u53ef\u4ee4\u52d5\u4f5c\u66f4\u5feb\u3001\u66f4\u5e73\u9806\u3002\u9019\u6bd4\u56fa\u5b9a\u6642\u9593\u540c\u6b65\u7684\u8a13\u7df4\u66f4\u8cbc\u8fd1\u771f\u5be6\u6a5f\u68b0\u4eba\u74b0\u5883\uff0c\u4f46\u6548\u679c\u4ecd\u53d6\u6c7a\u65bc\u611f\u6e2c\u5668\u983b\u7387\u3001\u63a7\u5236\u9031\u671f\u53ca\u4efb\u52d9\u672c\u8eab\uff0c\u4e0d\u80fd\u53ea\u4ee5\u5e40\u6578\u63a8\u65b7\u6240\u6709\u5834\u666f\u90fd\u6703\u6539\u5584\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u5efa\u57fa\u65bc openpi\uff0c\u4e26\u63d0\u4f9b JAX multi-node distributed training \u7684\u984d\u5916\u652f\u63f4\uff1bREADME \u540c\u6642\u5217\u51fa real-robot \u4efb\u52d9\u53ca\u793a\u7bc4\u5f71\u7247\u3002\u73fe\u6709\u8cc7\u6599\u672a\u63d0\u4f9b\u5b8c\u6574\u5b89\u88dd\u6d41\u7a0b\u3001\u53ef\u76f4\u63a5\u4e0b\u8f09\u7684\u6a21\u578b\u6b0a\u91cd\u6216\u6e2c\u8a66\u6307\u4ee4\uff0c\u5b98\u65b9\u9805\u76ee\u9801\u53ea\u8868\u793a\u7a0b\u5f0f\u78bc\u53ca\u6b0a\u91cd\u9810\u5b9a\u65bc 2026 \u5e74 8 \u6708 30 \u65e5\u516c\u958b\uff0c\u56e0\u6b64\u76ee\u524d\u8f03\u9069\u5408\u6a5f\u68b0\u4eba\u7814\u7a76\u5718\u968a\u53c3\u8003\u5176\u67b6\u69cb\u548c\u8a55\u6e2c\u65b9\u5411\uff0c\u800c\u975e\u7576\u4f5c\u5373\u6642\u53ef\u7528\u7684\u5957\u4ef6\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/happinesslz.github.io\/projects\/StreamPI\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/hku-sail\/StreamPI\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>StreamPI \u70ba\u55ae\u5e40 Vision-Language-Action \u6a21\u578b\u52a0\u5165\u8a18\u61b6\u8207\u5e7e\u4f55\u7dda\u7d22\uff0c\u4e26\u5c07\u591a\u5e40\u63a8\u7406\u7684\u5ef6\u9072\u589e\u5e45\u58d3\u81f3\u8f03\u4f4e\u6c34\u5e73\u3002<\/p>\n","protected":false},"author":8,"featured_media":11552,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,163,169,116,119,127,149,184,199,204],"tags":[],"class_list":["post-11553","post","type-post","status-publish","format-standard","hentry","category-133","category-163","category-169","category-agentic","category-119","category-127","category-149","category-robotic","category-dataset-","category-visionlanguageaction"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11553","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=11553"}],"version-history":[{"count":3,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11553\/revisions"}],"predecessor-version":[{"id":11557,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11553\/revisions\/11557"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11552"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11553"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11553"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11553"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}