{"id":78723,"date":"2026-04-06T08:12:41","date_gmt":"2026-04-06T01:12:41","guid":{"rendered":"https:\/\/itsystems.vn\/?p=78723"},"modified":"2026-07-29T11:25:02","modified_gmt":"2026-07-29T04:25:02","slug":"gemma-4-google-deepmind-ai-model","status":"publish","type":"post","link":"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/","title":{"rendered":"Gemma 4: Google&#8217;s DeepMind Releases the Pinnacle Open AI Model"},"content":{"rendered":"<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_86 counter-hierarchy ez-toc-counter ez-toc-light-blue ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">The content of the article<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Introduction_to_Gemma_4\" >Introduction to Gemma 4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Overview_of_Gemma_4\" >Overview of Gemma 4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Four_Standout_Versions_of_Gemma_4\" >Four Standout Versions of Gemma 4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Technical_Architecture\" >Technical Architecture<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Hybrid_Attention\" >Hybrid Attention<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Per-Layer_Embeddings_PLE\" >Per-Layer Embeddings (PLE)<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Shared_KV_Cache\" >Shared KV Cache<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Dual_RoPE_and_GQA\" >Dual RoPE and GQA<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Strengths_and_Standout_Capabilities_of_Gemma_4\" >Strengths and Standout Capabilities of Gemma 4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Simple_and_Effective_Guide_to_Running_Gemma_4\" >Simple and Effective Guide to Running Gemma 4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Frequently_Asked_Questions_about_Gemma_4\" >Frequently Asked Questions about Gemma 4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Related_services_from_IT_Systems\" >Related services from IT Systems<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-13\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#FAQ\" >FAQ<\/a><ul class='ez-toc-list-level-3' ><li class='ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-14\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#When_should_a_business_ask_IT_Systems_for_support\" >When should a business ask IT Systems for support?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-15\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Can_IT_Systems_help_review_the_current_environment_before_proposing_a_solution\" >Can IT Systems help review the current environment before proposing a solution?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-16\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Does_this_topic_connect_to_ongoing_IT_operations\" >Does this topic connect to ongoing IT operations?<\/a><\/li><\/ul><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-17\" href=\"https:\/\/itsystems.vn\/en\/gemma-4-google-deepmind-ai-model\/#Need_help_applying_this_to_your_business\" >Need help applying this to your business?<\/a><\/li><\/ul><\/nav><\/div>\n<h2><span class=\"ez-toc-section\" id=\"Introduction_to_Gemma_4\"><\/span>Introduction to Gemma 4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Gemma 4 marks a breakthrough in the open-source AI field as Google DeepMind announced this model on April 2, 2026. This is the latest generation in Google&#8217;s open AI product line, built on advanced technology from Gemini 3. With four diverse versions from 2B to 31B parameters, <strong>Gemma 4<\/strong> runs smoothly on any device, from mobile phones to powerful servers. For the first time, Google applies a fully open Apache 2.0 license, enabling the model to quickly achieve over 400 million downloads and thousands of custom variants in a short time. Its performance surpasses Gemma 3 and even competes head-to-head with heavyweights like Qwen 3.5 or Llama 4 on many standard benchmarks.<\/p>\n<p>In the context of rapidly developing AI, <strong>Gemma 4<\/strong> not only delivers superior computational power but also emphasizes accessibility. Developers can easily integrate it into real-world applications, from personalized chatbots to intelligent image analysis systems. For example, a tech startup can use the small version to run on edge devices, significantly saving cloud costs compared to closed models like the GPT series.<\/p>\n<p><a href=\"https:\/\/azdigi.com\/blog\/cong-nghe\/gemma-4-la-gi-model-ai-mo-manh-nhat-cua-google-chay-tu-dien-thoai-den-server\" target=\"_blank\" rel=\"nofollow noopener\"><br \/>\nReference source<br \/>\n<\/a><\/p>\n<h2><span class=\"ez-toc-section\" id=\"Overview_of_Gemma_4\"><\/span>Overview of Gemma 4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>Gemma 4<\/strong> is designed with four model sizes suitable for every need, from mobile devices to specialized workstations. These versions support multimodality (<strong>Gemma 4 multimodal<\/strong>), naturally handling text and images, while some variants extend to audio and video. The standout feature is the Apache 2.0 license, allowing commercial use without any restrictions, a significant difference from the previous Gemma 3.<\/p>\n<p>The core architecture inherits from Gemini 3, praised by Google CEO Sundar Pichai and DeepMind CEO Demis Hassabis as &#8220;the world&#8217;s best open models at their size.&#8221; This means <strong>Gemma 4<\/strong> achieves high performance without requiring massive resources. In practice, a programmer can deploy it on a personal laptop to build a virtual assistant for real-time data analysis, such as object recognition in surveillance camera videos.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Four_Standout_Versions_of_Gemma_4\"><\/span>Four Standout Versions of Gemma 4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>Gemma 4<\/strong> offers diversity with four optimized versions:<\/p>\n<ul>\n<li><strong>E2B<\/strong>: With 5.1B parameters (effective 2.3B), supports 128K token context, handles text, images, and audio. Ideal for lightweight mobile apps like real-time language translation combined with speech recognition.<\/li>\n<li><strong>E4B<\/strong>: 8B parameters (effective 4.5B), 128K context, extends to video alongside text, images, and audio. Perfect for <strong>Gemma 4 multimodal<\/strong> projects like multimedia content analysis on edge devices.<\/li>\n<li><strong>26B MoE (A4B)<\/strong>: 25.2B parameters (active 3.8B), 256K context, focuses on text and images. Uses Mixture of Experts architecture for faster inference, ideal for workstations handling long documents.<\/li>\n<li><strong>31B Dense<\/strong>: 30.7B parameters, 256K context, supports text and images. This is the most powerful version for complex tasks like content creation or scientific research.<\/li>\n<\/ul>\n<p>Each version is fine-tuned to balance performance and resources. For instance, the E2B version can run on Android smartphones with limited RAM, allowing individual users to experience high-end AI without an internet connection.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Technical_Architecture\"><\/span>Technical Architecture<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>\nGemma 4 introduces several architectural improvements compared to traditional transformer models.<br \/>\nThese enhancements are designed to improve efficiency while maintaining strong performance when processing long contexts.\n<\/p>\n<h3><span class=\"ez-toc-section\" id=\"Hybrid_Attention\"><\/span>Hybrid Attention<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\nInstead of relying entirely on global attention &#8211; which can consume significant memory when dealing with long contexts &#8211;<br \/>\nGemma 4 combines two attention mechanisms. The model alternates between<br \/>\n<strong>sliding-window attention<\/strong>, which focuses on the most recent 512 or 1024 tokens,<br \/>\nand <strong>global full-context attention<\/strong>.\n<\/p>\n<p>\nAt the final layer, the model always applies global attention so it can still consider the entire context when necessary.<br \/>\nThis hybrid approach significantly reduces memory usage while maintaining the ability to understand long sequences.\n<\/p>\n<h3><span class=\"ez-toc-section\" id=\"Per-Layer_Embeddings_PLE\"><\/span>Per-Layer Embeddings (PLE)<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\nIn many transformer architectures, all layers share the same input embedding representation.<br \/>\nGemma 4 introduces a different approach by providing each layer with its own conditioning vector.\n<\/p>\n<p>\nThis means every layer receives an additional signal that helps define its role within the network.<br \/>\nAs a result, the model can learn more effectively without significantly increasing the number of parameters.\n<\/p>\n<h3><span class=\"ez-toc-section\" id=\"Shared_KV_Cache\"><\/span>Shared KV Cache<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\nIn the later layers of the model, Gemma 4 reuses the <strong>Key\/Value cache<\/strong> from previous layers<br \/>\ninstead of computing it independently for every layer.\n<\/p>\n<p>\nIn simpler terms, some layers can borrow KV cache data that has already been generated earlier in the network.<br \/>\nThis reduces both memory consumption and computational cost while keeping the output quality nearly unchanged.\n<\/p>\n<h3><span class=\"ez-toc-section\" id=\"Dual_RoPE_and_GQA\"><\/span>Dual RoPE and GQA<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>\nGemma 4 uses two different base frequencies for its<br \/>\n<strong>Rotary Position Embedding (RoPE)<\/strong>.<br \/>\nSliding-window layers typically use a base of around <strong>10K<\/strong>,<br \/>\nwhile global layers use a much larger base of around <strong>1M<\/strong>.\n<\/p>\n<p>\nThe larger frequency allows global layers to represent positional information more accurately when handling long contexts.\n<\/p>\n<p>\nAdditionally, <strong>Grouped Query Attention (GQA)<\/strong> is configured differently for local and global layers.<br \/>\nLocal layers usually use around 2 queries per KV head, while global layers may use up to 8 queries.<br \/>\nThis design helps optimize memory usage while allowing global layers to process the entire context more effectively.\n<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Strengths_and_Standout_Capabilities_of_Gemma_4\"><\/span>Strengths and Standout Capabilities of Gemma 4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>One of the breakthrough features of <strong>Gemma 4<\/strong> is chain-of-thought reasoning support via the special token &lt;|think|&gt;, enabling deeper logical problem analysis. In tests like GPQA, it achieves 85.7% on reasoning tasks, on par with Qwen 3.5 27B, excelling in <strong>Gemma 4 multimodal<\/strong> capabilities and open licensing, despite a slightly smaller context window.<\/p>\n<p>Shared KV Cache technology is another highlight, allowing cache reuse across layers, significantly reducing memory and computation time. Real-world example: In an enterprise chatbot system, this feature handles thousands of simultaneous queries without increasing server load. Compared to competitors, <strong>Gemma 4<\/strong> stands out in local deployment, better data security, and lower costs for small companies.<\/p>\n<p>Moreover, the model supports easy fine-tuning, allowing customization for specific fields like healthcare (X-ray image analysis) or education (interactive lesson creation). Benchmarks show it surpasses Gemma 3 in inference speed by up to 30% and higher multilingual accuracy.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Simple_and_Effective_Guide_to_Running_Gemma_4\"><\/span>Simple and Effective Guide to Running Gemma 4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Deploying <strong>Gemma 4<\/strong> has never been easier. With Ollama &#8211; a popular tool for local AI &#8211; you just need the command <code>ollama run gemma-4<\/code> to download and run it immediately. Check the model list with <code>ollama list<\/code>. This is ideal for beginners, especially with <strong>Gemma 4 Ollama<\/strong> on personal computers.<\/p>\n<p>For powerful servers, use Llama-server: <code>.\/llama-server -m gemma-4-26b-a4b-Q4_K_M.gguf -c 8192 -ngl 99<\/code>. On macOS with MLX, install via <code>pip install mlx-lm<\/code> then run <code>mlx_lm.generate --model google\/gemma-4-26b-a4b-mlx --prompt \"Hello\"<\/code>. These guides have been successfully used by millions of developers, from building Telegram bots to integrating into web apps.<\/p>\n<p>Note: Ensure GPU support for optimal speed. For example, on Ubuntu VPS, combine Ollama with Docker for easy scaling, suitable for production.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Frequently_Asked_Questions_about_Gemma_4\"><\/span>Frequently Asked Questions about Gemma 4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>How is Gemma 4 different from Gemini?<\/strong> Gemma 4 is an open-weight model that runs completely locally without cloud, while Gemini is closed-source based on Google&#8217;s cloud services.<\/p>\n<ul>\n<li><strong>Can it be used commercially?<\/strong> Yes, thanks to Apache 2.0.<\/li>\n<li><strong>Hardware requirements?<\/strong> From basic CPU to high-end GPU depending on the version.<\/li>\n<li><strong>Integration with which frameworks?<\/strong> Hugging Face, Ollama, MLX all support it well.<\/li>\n<\/ul>\n<p>For more information, refer to guides on installing Ollama on Ubuntu VPS or articles on fine-tuning <strong>Gemma 4<\/strong>. With continuous updates, Gemma 4 promises to shape the future of open-source AI.<\/p>\n<p><!-- its-deep-aio-en-related-2026-07-27 --><\/p>\n<section class=\"its-deep-aio-related\" aria-label=\"Related IT Systems services\">\n<h2><span class=\"ez-toc-section\" id=\"Related_services_from_IT_Systems\"><\/span>Related services from IT Systems<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>If your team is dealing with this issue in a live business environment, these services can help you move from diagnosis to a stable operating process.<\/p>\n<ul>\n<li><a href=\"https:\/\/itsystems.vn\/en\/microsoft-business-licensing\/microsoft-windows-licensing\/\">Windows licensing for business<\/a><\/li>\n<li><a href=\"https:\/\/itsystems.vn\/en\/microsoft-business-licensing\/windows-license-pricing\/\">Windows license pricing<\/a><\/li>\n<li><a href=\"https:\/\/itsystems.vn\/en\/it-services-for-businesses\/it-support-services\/\">IT support services<\/a><\/li>\n<\/ul>\n<\/section>\n<p><!-- its-deep-aio-en-faq-2026-07-27 --><\/p>\n<section class=\"its-deep-aio-faq\" aria-label=\"Frequently asked questions\">\n<h2><span class=\"ez-toc-section\" id=\"FAQ\"><\/span>FAQ<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<h3><span class=\"ez-toc-section\" id=\"When_should_a_business_ask_IT_Systems_for_support\"><\/span>When should a business ask IT Systems for support?<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>Ask for support when the issue affects users, business data, security, licensing compliance, service availability or daily operations. A short technical review often prevents repeated incidents and hidden costs.<\/p>\n<h3><span class=\"ez-toc-section\" id=\"Can_IT_Systems_help_review_the_current_environment_before_proposing_a_solution\"><\/span>Can IT Systems help review the current environment before proposing a solution?<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>Yes. IT Systems can review the current setup, identify risks, map the issue to the right service scope and recommend a practical next step for your business.<\/p>\n<h3><span class=\"ez-toc-section\" id=\"Does_this_topic_connect_to_ongoing_IT_operations\"><\/span>Does this topic connect to ongoing IT operations?<span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p>In most cases, yes. Problems around software, cloud, endpoint, network, backup or security should be connected to a broader IT operations plan instead of being handled as isolated incidents.<\/p>\n<\/section>\n<p><!-- its-deep-aio-en-cta-2026-07-27 --><\/p>\n<section class=\"its-deep-aio-cta\" aria-label=\"Contact IT Systems\">\n<h2><span class=\"ez-toc-section\" id=\"Need_help_applying_this_to_your_business\"><\/span>Need help applying this to your business?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>IT Systems Vietnam can help assess the issue, recommend the right service path and support implementation for your team.<\/p>\n<p><a class=\"button\" href=\"https:\/\/itsystems.vn\/en\/contact-it-systems-vietnam\/\">Contact IT Systems<\/a> <a class=\"button\" href=\"https:\/\/itsystems.vn\/en\/it-services-for-businesses\/it-support-services\/\">View IT support services<\/a><\/p>\n<\/section>\n","protected":false},"excerpt":{"rendered":"<p>Introduction to Gemma 4 Gemma 4 marks a breakthrough in the open-source AI field as Google DeepMind announced this model on April 2, 2026. This is the latest generation in Google&#8217;s open AI product line, built on advanced technology from Gemini 3. With four diverse versions from 2B to 31B parameters, Gemma 4 runs smoothly [&hellip;]<\/p>\n","protected":false},"author":54,"featured_media":78721,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_focus_keyword":"Gemma 4,Gemma 4 Ollama, Gemma 4 multimodal","rank_math_title":"","rank_math_description":"Gemma 4 is a powerful open AI model from Google DeepMind with 4 multimodal versions, Apache 2.0 license. Guide to running Gemma 4 Ollama, performance comparison, and real-world applications.","rank_math_robots":"","rank_math_canonical_url":"","rank_math_schema":"","footnotes":""},"categories":[1344],"tags":[],"class_list":["post-78723","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized-en"],"wpml_current_locale":"en_US","wpml_translations":{"vi_VN":{"locale":"vi_VN","id":78720,"slug":"gemma-4-model-ai-google-deepmind","post_title":"Gemma 4: Model AI M\u1edf \u0110\u1ec9nh Cao T\u1eeb Google DeepMind \u0110\u00e3 Ch\u00ednh Th\u1ee9c Ra M\u1eaft","href":"https:\/\/itsystems.vn\/gemma-4-model-ai-google-deepmind\/"}},"_links":{"self":[{"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/posts\/78723","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/users\/54"}],"replies":[{"embeddable":true,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/comments?post=78723"}],"version-history":[{"count":2,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/posts\/78723\/revisions"}],"predecessor-version":[{"id":85862,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/posts\/78723\/revisions\/85862"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/media\/78721"}],"wp:attachment":[{"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/media?parent=78723"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/categories?post=78723"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itsystems.vn\/en\/wp-json\/wp\/v2\/tags?post=78723"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}