{"id":26828,"date":"2025-04-17T06:34:09","date_gmt":"2025-04-17T06:34:09","guid":{"rendered":"https:\/\/www.europesays.com\/uk\/26828\/"},"modified":"2025-04-17T06:34:09","modified_gmt":"2025-04-17T06:34:09","slug":"microsoft-researchers-say-theyve-developed-a-hyper-efficient-ai-model-that-can-run-on-cpus","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/uk\/26828\/","title":{"rendered":"Microsoft researchers say they&#8217;ve developed a hyper-efficient AI model that can run on CPUs"},"content":{"rendered":"<p id=\"speakable-summary\" class=\"wp-block-paragraph\">Microsoft researchers claim they\u2019ve developed the largest-scale 1-bit AI model, also known as a \u201cbitnet,\u201d to date. Called BitNet b1.58 2B4T, it\u2019s <a href=\"https:\/\/huggingface.co\/microsoft\/bitnet-b1.58-2B-4T\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">openly available<\/a> under an MIT license and can run on CPUs, including Apple\u2019s M2.<\/p>\n<p class=\"wp-block-paragraph\">Bitnets are essentially compressed models designed to run on lightweight hardware. In standard models, weights, the values that define the internal structure of a model, are often quantized so the models perform well on a wide range of machines. Quantizing the weights lowers the number of bits \u2014 the smallest units a computer can process \u2014 needed to represent those weights, enabling models to run on chips with less memory, faster.<\/p>\n<p class=\"wp-block-paragraph\">Bitnets quantize weights into just three values: -1, 0, and 1. In theory, that makes them far more memory- and computing-efficient than most models today.<\/p>\n<p class=\"wp-block-paragraph\">The Microsoft researchers say that BitNet b1.58 2B4T is the first bitnet with 2 billion parameters, \u201cparameters\u201d being largely synonymous with \u201cweights.\u201d Trained on a dataset of 4 trillion tokens \u2014 equivalent to about 33 million books, <a href=\"https:\/\/everdome.io\/news\/ai-tokens-explained-the-tldr-version\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">by one estimate<\/a> \u2014 BitNet b1.58 2B4T outperforms traditional models of similar sizes, the researchers claim.<\/p>\n<p class=\"wp-block-paragraph\">BitNet b1.58 2B4T doesn\u2019t sweep the floor with rival 2 billion-parameter models, to be clear, but it seemingly holds its own. According to the researchers\u2019 testing, the model surpasses Meta\u2019s Llama 3.2 1B, Google\u2019s Gemma 3 1B, and Alibaba\u2019s Qwen 2.5 1.5B on benchmarks including GSM8K (a collection of grade-school-level math problems) and PIQA (which tests physical commonsense reasoning skills).<\/p>\n<p class=\"wp-block-paragraph\">Perhaps more impressively, BitNet b1.58 2B4T is speedier than other models of its size \u2014 in some cases, twice the speed \u2014 while using a fraction of the memory. <\/p>\n<p class=\"wp-block-paragraph\">There is a catch, however. <\/p>\n<p class=\"wp-block-paragraph\">Achieving that performance requires using Microsoft\u2019s custom framework, bitnet.cpp, which only works with certain hardware at the moment. Absent from the list of supported chips are GPUs, which dominate the AI infrastructure landscape. <\/p>\n<p class=\"wp-block-paragraph\">That\u2019s all to say that bitnets may hold promise, particularly for resource-constrained devices. But compatibility is \u2014 and will likely remain \u2014 a big sticking point.<\/p>\n","protected":false},"excerpt":{"rendered":"Microsoft researchers claim they\u2019ve developed the largest-scale 1-bit AI model, also known as a \u201cbitnet,\u201d to date. Called&hellip;\n","protected":false},"author":2,"featured_media":26829,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3163],"tags":[323,1942,16292,507,53,16,15],"class_list":{"0":"post-26828","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-artificial-intelligence","10":"tag-bitnet","11":"tag-microsoft","12":"tag-technology","13":"tag-uk","14":"tag-united-kingdom"},"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@uk\/114351915842732303","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/posts\/26828","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/comments?post=26828"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/posts\/26828\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/media\/26829"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/media?parent=26828"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/categories?post=26828"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/tags?post=26828"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}