{"id":16157,"date":"2026-11-17T11:04:34","date_gmt":"2026-11-17T02:04:34","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/?p=16157"},"modified":"2026-11-17T11:14:00","modified_gmt":"2026-11-17T02:14:00","slug":"duo-1-0-distill-quant-fp8-realization-weights","status":"publish","type":"post","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/duo-1-0-distill-quant-fp8-realization-weights\/","title":{"rendered":"Duo-1.0-distill-quant FP8 Quantized Realization Model: 80GB+ VRAM"},"content":{"rendered":"<p><!-- lmw:facts --><\/p>\n<h2>At a Glance<\/h2>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Value<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Repository<\/td>\n<td><a href=\"https:\/\/huggingface.co\/inclusionAI\/Duo-1.0-distill-quant\">inclusionAI\/Duo-1.0-distill-quant<\/a><\/td>\n<\/tr>\n<tr>\n<td>Publisher guide<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/publisher-inclusionai-en\/\">inclusionAI (Ant Group): models and licenses<\/a><\/td>\n<\/tr>\n<tr>\n<td>Published<\/td>\n<td>2026-11-13<\/td>\n<\/tr>\n<tr>\n<td>License<\/td>\n<td>minimax-h3-community-license<\/td>\n<\/tr>\n<tr>\n<td>Formats<\/td>\n<td><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-safetensors-en\/\">safetensors<\/a><\/td>\n<\/tr>\n<tr>\n<td>Paper<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2611.15131\">arXiv:2611.15131<\/a><\/td>\n<\/tr>\n<tr>\n<td>Source type<\/td>\n<td>Primary source (the publisher itself)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Values determined by this site&#8217;s code when the information was collected. Dates are JST.<\/em><\/p>\n<p><!-- \/lmw:facts --><\/p>\n<h2>Overview<\/h2>\n<p>inclusionAI has released &#8220;Duo-1.0-distill-quant&#8221;, the Realization weight package for &#8220;Duo 1.0&#8221;, a world model that enables real-time full-duplex multimodal interaction. This repository provides 4-step distilled, FP8 quantized weights designed for the Realization module of Duo 1.0. When in actual use, these weights are overlaid and loaded into the directory of the base package, <code>inclusionAI\/Duo-1.0-causal-base<\/code>.<\/p>\n<p>Details regarding the background research of the model are published in the paper &#8220;<a href=\"https:\/\/arxiv.org\/abs\/2611.15131\">Duo: A Reciprocal World Model for Full-Duplex Interaction<\/a>&#8220;. This paper introduces a novel concept called the &#8220;Reciprocal World Model (RWM)&#8221;, which treats intelligence and physical reality as equals. While many conventional world models tend to focus on simulating physical reality and treat reasoning and decision-making as external controls, RWM positions reasoning and cognition as essential components that constitute the world itself.<\/p>\n<p>The RWM architecture defines two core capabilities. The first is the &#8220;Intent&#8221; module, which handles cognition and decision-making in interaction with the world. The second is the &#8220;Realization&#8221; module, which generates coherent audio and video behaviors in real-time. To achieve this, Duo v1 incorporates three key technical designs.<\/p>\n<p>First, it utilizes autoregressive modeling not just for instruction understanding in diffusion models, but for continuous conversational reasoning and participation decisions in interactions.<\/p>\n<p>Second, it synchronizes Intent and Realization tokens on a unified timeline to construct chunk-specific conditions. By applying module-specific attention masks, it maintains causal dependencies within and across chunks, preventing access to future input data.<\/p>\n<p>Third, it employs dual rollout distillation to suppress distribution shift in both modules. Training is performed by combining autoregressively generated Intent tokens with the synthetic audiovisual history generated by Realization.<\/p>\n<p>Through these designs, Duo v1 achieves full-duplex interaction that supports natural conversation and user interruptions while continuously sensing audio, video, and text inputs in real-time. It supports diverse use cases without requiring keyboard or mouse operation, such as sharing physical objects, modifying virtual space configurations, assistant guidance, and playing audiovisual games.<\/p>\n<h2>Specifications<\/h2>\n<ul>\n<li><strong>Version<\/strong>: Duo 1.0<\/li>\n<li><strong>Variant<\/strong>: FP8 quantized 4-step distilled Realization<\/li>\n<li><strong>Dependent Base Model<\/strong>: <code>inclusionAI\/Duo-1.0-causal-base<\/code><\/li>\n<li><strong>Realization Format<\/strong>: ModelOpt FP8 (per-channel scaling)<\/li>\n<li><strong>Default Sampler<\/strong>: Euler (4 steps, flow shift 5.0)<\/li>\n<\/ul>\n<h2>Strengths and Use Cases<\/h2>\n<p>Duo-1.0-distill-quant excels at full-duplex multimodal interaction, capable of continuously perceiving multiple modalities\u2014audio, video, and text\u2014in real-time and returning simultaneous responses. This model features video-generation and audio-generation capabilities, making it suitable for building systems accompanied by real-time natural dialogue and reactions.<\/p>\n<p>The specific intended uses and use cases gathered from the documentation are as follows:<\/p>\n<ul>\n<li><strong>Natural Conversation and User Interruption Handling<\/strong>: Because it continuously processes audio and text inputs in real-time, it can flexibly and smoothly respond even if a user interrupts speech in the middle of a conversation.<\/li>\n<li><strong>Physical Object Sharing and Assistant Functions via Camera Images<\/strong>: It supports use cases where users explain things while sharing physical objects at hand through a camera, or give real-time instructions and guidance to an AI assistant.<\/li>\n<li><strong>Virtual Space Operations and Gaming without Mouse or Keyboard<\/strong>: It realizes interactive environments where users can modify virtual scene configurations or play audiovisual games through voice and visual operations without using traditional input devices.<\/li>\n<\/ul>\n<p>Thus, by combining the Intent module for cognition and the Realization module for generation and expression, it can be utilized as an advanced world model that goes beyond mere media generation.<\/p>\n<p><!-- lmw:hardware --><\/p>\n<h2>Hardware Requirements<\/h2>\n<p><strong>Estimated requirements (calculated by Local Model Watch)<\/strong><\/p>\n<div class=\"lmw-table-scroll\" tabindex=\"0\" style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;max-width:100%;\">\n<table style=\"width:max-content;min-width:100%;border-collapse:collapse;\">\n<thead>\n<tr>\n<th>Your VRAM<\/th>\n<th>Quantization<\/th>\n<th>File size<\/th>\n<th>Est. memory needed<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>80GB class (A100 \/ H100)<\/td>\n<td>Original precision(realization)<\/td>\n<td>43.8GB<\/td>\n<td>52.6GB<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<p><em>Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model&#8217;s authors. Compare with other models in our <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/\">VRAM quick reference<\/a>. What the quantization names mean: <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/glossary-quantization-en\/\">glossary<\/a>.<\/em><\/p>\n<p><!-- \/lmw:hardware --><\/p>\n<p><!-- lmw:runnability --><\/p>\n<h2>Can You Run It Locally?<\/h2>\n<p>The publisher distributes this model as safetensors.<\/p>\n<p><strong>License \u2014 <code>minimax-h3-community-license<\/code>:<\/strong> A custom license from the publisher. Check the original terms directly, including whether commercial use is permitted.<\/p>\n<p><em>Compiled by this site&#8217;s code from the published formats and the license field. License summaries are not legal advice \u2014 check the publisher&#8217;s original terms before relying on them.<\/em><\/p>\n<p><!-- \/lmw:runnability --><\/p>\n<p><!-- lmw:lab --><\/p>\n<h2>Our Own Measurements<\/h2>\n<h3>Measurements We Did Not Take<\/h3>\n<p>We have not confirmed that this site meets the commercial-use terms of this model&#8217;s license (<code>minimax-h3-community-license<\/code>). Because this site carries advertising, we did not run the model (no CPU run, answers, quantization comparison, conversion or generation).<\/p>\n<p><!-- \/lmw:lab --><\/p>\n<h2>How to Get It<\/h2>\n<p>Duo-1.0-distill-quant is available on Hugging Face in safetensors format. The format employs ModelOpt FP8 (per-channel scaling).<\/p>\n<p>To use this model, you must first download <code>inclusionAI\/Duo-1.0-causal-base<\/code>, which contains the base configuration, and place the Realization weights of this repository (<code>inclusionAI\/Duo-1.0-distill-quant<\/code>) by overwriting (overlaying) them into the same destination directory. The shared Intent module, various codecs, audio models, and the Realization backbone are provided by the base package side.<\/p>\n<p>The official documentation guides the model acquisition using the following download procedure with the <code>hf<\/code> command or similar:<\/p>\n<pre><code class=\"language-bash\">hf download inclusionAI\/Duo-1.0-causal-base --local-dir weights\nhf download inclusionAI\/Duo-1.0-distill-quant --local-dir weights\n<\/code><\/pre>\n<p>At runtime, the runtime you use must be compatible with the packaged ModelOpt FP8 format or an equivalent conversion path. For details such as implementation code, configuration files, and integration methods into various runtimes, please refer to the official GitHub repository <a href=\"https:\/\/github.com\/inclusionAI\/Duo\">inclusionAI\/Duo<\/a>.<\/p>\n<p><!-- lmw:related --><\/p>\n<h2>Related Articles<\/h2>\n<ul>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/duo-1-0-real-time-audio-visual-world-models\/\">Duo-1.0: Real-Time Audio-Visual Full-Duplex World Models<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/ling-3-0-flash-sante-fp8-released\/\">Ling-3.0-flash-Sante-fp8 Text Generation Model: ~144GB Memory<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/duo-12-causal-base-released\/\">Duo-1.2-causal-base Released for Real-time Multimodal Models<\/a><\/li>\n<li><a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/2026\/11\/17\/inclusionai-sing-vl-s1-4b-gguf\/\">SingVL_S1_4B_GGUF 4B Multimodal Decision-Making Model: 4GB+ VRAM<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:related --><\/p>\n<p><!-- lmw:next-steps --><\/p>\n<h2>What to Read Next<\/h2>\n<ul>\n<li><strong>Find models by VRAM<\/strong> (This model needs at least 80GB) \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/\">VRAM quick reference<\/a><\/li>\n<li><strong>Formats this model is available in<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/format-safetensors-en\/\">Safetensors format guide and models<\/a><\/li>\n<li><strong>Learn about the publisher<\/strong> \u2192 <a href=\"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/publisher-inclusionai-en\/\">inclusionAI (Ant Group): models, licenses and articles<\/a><\/li>\n<\/ul>\n<p><!-- \/lmw:next-steps --><\/p>\n<h2>Sources<\/h2>\n<ul>\n<li><a href=\"https:\/\/huggingface.co\/inclusionAI\/Duo-1.0-distill-quant\">https:\/\/huggingface.co\/inclusionAI\/Duo-1.0-distill-quant<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2611.15131\">https:\/\/arxiv.org\/abs\/2611.15131<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/inclusionAI\/Duo\">https:\/\/github.com\/inclusionAI\/Duo<\/a><\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Discover Duo-1.0-distill-quant, an FP8 4-step distilled Realization weight for inclusionAI&#8217;s Duo 1.0 real-time full-duplex world model.<\/p>\n","protected":false},"author":1,"featured_media":16156,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[310],"tags":[4991,5016,1133,1691,1547,896,5018],"class_list":["post-16157","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-new-models","tag-duo-1-0-en","tag-duo-1-0-quant-en","tag-fp8-en","tag-inclusionai-en","tag-verified","tag--en"],"lang":"en","translations":{"en":16157,"ja":16155},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/16157","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=16157"}],"version-history":[{"count":2,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/16157\/revisions"}],"predecessor-version":[{"id":18120,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/posts\/16157\/revisions\/18120"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media\/16156"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=16157"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/categories?post=16157"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/tags?post=16157"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}