{"id":201,"date":"2024-07-27T11:10:29","date_gmt":"2024-07-27T11:10:29","guid":{"rendered":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/"},"modified":"2024-07-27T11:10:29","modified_gmt":"2024-07-27T11:10:29","slug":"flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage","status":"publish","type":"post","link":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/","title":{"rendered":"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l&rsquo;inf\u00e9rence des grands mod\u00e8les de langage"},"content":{"rendered":"<p><img decoding=\"async\" src=\"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&#038;se=2024-07-27T12%3A10%3A28Z&#038;sp=r&#038;sv=2023-11-03&#038;sr=b&#038;rscd=inline&#038;rsct=image\/png&#038;skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&#038;sktid=a48cca56-e6da-484e-a814-9c849652bcb3&#038;skt=2024-07-26T23%3A27%3A09Z&#038;ske=2024-07-27T23%3A27%3A09Z&#038;sks=b&#038;skv=2023-11-03&#038;sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D\"><\/p>\n<p>Les mod\u00e8les de langage \u00e0 grande \u00e9chelle (LLM) rencontrent des d\u00e9fis de d\u00e9ploiement dus \u00e0 la latence caus\u00e9e par les contraintes de bande passante m\u00e9moire. Les chercheurs utilisent la quantification des poids uniquement pour r\u00e9soudre ce probl\u00e8me, en compressant les param\u00e8tres des LLM \u00e0 une pr\u00e9cision plus faible. Cette m\u00e9thode r\u00e9duit la latence et les besoins en m\u00e9moire GPU. La mise en \u0153uvre efficace requiert des noyaux de multiplication de matrices sur mesure, capables de d\u00e9placer, d\u00e9quantifier et traiter les poids efficacement. FLUTE, une approche innovante, traite les matrices sous 8 bits, optimise la d\u00e9quantification bas\u00e9e sur des tables de correspondance et am\u00e9liore la r\u00e9partition des charges de travail. Avec la restructuration des poids hors ligne, des tables de recherche vectoris\u00e9es en m\u00e9moire partag\u00e9e, et le partitionnement Stream-K, FLUTE g\u00e8re la quantification faible en bits et non uniforme, am\u00e9liorant performance et efficacit\u00e9. Ses performances sont examin\u00e9es sur des GPUs A6000 et A100, montrant des acc\u00e9l\u00e9rations significatives et une flexibilit\u00e9 dans diverses configurations de quantification.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Les mod\u00e8les de langage \u00e0 grande \u00e9chelle (LLM) rencontrent des d\u00e9fis de d\u00e9ploiement dus \u00e0 la latence caus\u00e9e par les contraintes de bande passante m\u00e9moire. Les chercheurs utilisent la quantification des poids uniquement pour r\u00e9soudre ce probl\u00e8me, en compressant les param\u00e8tres des LLM \u00e0 une pr\u00e9cision plus faible. Cette m\u00e9thode r\u00e9duit la latence et les [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-201","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v25.5 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l&#039;inf\u00e9rence des grands mod\u00e8les de langage - IA &amp; TECHNOLOGIES<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l&#039;inf\u00e9rence des grands mod\u00e8les de langage - IA &amp; TECHNOLOGIES\" \/>\n<meta property=\"og:description\" content=\"Les mod\u00e8les de langage \u00e0 grande \u00e9chelle (LLM) rencontrent des d\u00e9fis de d\u00e9ploiement dus \u00e0 la latence caus\u00e9e par les contraintes de bande passante m\u00e9moire. Les chercheurs utilisent la quantification des poids uniquement pour r\u00e9soudre ce probl\u00e8me, en compressant les param\u00e8tres des LLM \u00e0 une pr\u00e9cision plus faible. Cette m\u00e9thode r\u00e9duit la latence et les [&hellip;]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/\" \/>\n<meta property=\"og:site_name\" content=\"IA &amp; TECHNOLOGIES\" \/>\n<meta property=\"article:published_time\" content=\"2024-07-27T11:10:29+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D\" \/>\n<meta name=\"author\" content=\"m\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"m\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"1 minute\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/\",\"url\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/\",\"name\":\"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l'inf\u00e9rence des grands mod\u00e8les de langage - IA &amp; TECHNOLOGIES\",\"isPartOf\":{\"@id\":\"https:\/\/www.asm15.fr\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D\",\"datePublished\":\"2024-07-27T11:10:29+00:00\",\"author\":{\"@id\":\"https:\/\/www.asm15.fr\/#\/schema\/person\/9e2a0039267955de05a7ce4282a905e6\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#primaryimage\",\"url\":\"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D\",\"contentUrl\":\"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Accueil\",\"item\":\"https:\/\/www.asm15.fr\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l&rsquo;inf\u00e9rence des grands mod\u00e8les de langage\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.asm15.fr\/#website\",\"url\":\"https:\/\/www.asm15.fr\/\",\"name\":\"IA & TECHNOLOGIES\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.asm15.fr\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.asm15.fr\/#\/schema\/person\/9e2a0039267955de05a7ce4282a905e6\",\"name\":\"m\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/www.asm15.fr\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/6fe5cecb08501f51d81efa38bdd65daa775b9224f6dc9617c8615be766310093?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/6fe5cecb08501f51d81efa38bdd65daa775b9224f6dc9617c8615be766310093?s=96&d=mm&r=g\",\"caption\":\"m\"},\"sameAs\":[\"https:\/\/www.asm15.fr\"],\"url\":\"https:\/\/www.asm15.fr\/index.php\/author\/m\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l'inf\u00e9rence des grands mod\u00e8les de langage - IA &amp; TECHNOLOGIES","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/","og_locale":"fr_FR","og_type":"article","og_title":"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l'inf\u00e9rence des grands mod\u00e8les de langage - IA &amp; TECHNOLOGIES","og_description":"Les mod\u00e8les de langage \u00e0 grande \u00e9chelle (LLM) rencontrent des d\u00e9fis de d\u00e9ploiement dus \u00e0 la latence caus\u00e9e par les contraintes de bande passante m\u00e9moire. Les chercheurs utilisent la quantification des poids uniquement pour r\u00e9soudre ce probl\u00e8me, en compressant les param\u00e8tres des LLM \u00e0 une pr\u00e9cision plus faible. Cette m\u00e9thode r\u00e9duit la latence et les [&hellip;]","og_url":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/","og_site_name":"IA &amp; TECHNOLOGIES","article_published_time":"2024-07-27T11:10:29+00:00","og_image":[{"url":"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D","type":"","width":"","height":""}],"author":"m","twitter_card":"summary_large_image","twitter_misc":{"\u00c9crit par":"m","Dur\u00e9e de lecture estim\u00e9e":"1 minute"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/","url":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/","name":"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l'inf\u00e9rence des grands mod\u00e8les de langage - IA &amp; TECHNOLOGIES","isPartOf":{"@id":"https:\/\/www.asm15.fr\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#primaryimage"},"image":{"@id":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#primaryimage"},"thumbnailUrl":"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D","datePublished":"2024-07-27T11:10:29+00:00","author":{"@id":"https:\/\/www.asm15.fr\/#\/schema\/person\/9e2a0039267955de05a7ce4282a905e6"},"breadcrumb":{"@id":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#primaryimage","url":"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D","contentUrl":"https:\/\/oaidalleapiprodscus.blob.core.windows.net\/private\/org-PXteLUy4B6mslwRon6h7ri9z\/user-0jfma6aPNvZyLXZUbw4CmvZG\/img-4CIW58yg3Kehyg5oqPnTq67D.png?st=2024-07-27T10%3A10%3A28Z&se=2024-07-27T12%3A10%3A28Z&sp=r&sv=2023-11-03&sr=b&rscd=inline&rsct=image\/png&skoid=6aaadede-4fb3-4698-a8f6-684d7786b067&sktid=a48cca56-e6da-484e-a814-9c849652bcb3&skt=2024-07-26T23%3A27%3A09Z&ske=2024-07-27T23%3A27%3A09Z&sks=b&skv=2023-11-03&sig=lLa\/54vunm29vwHE%2Bv1VrKRS\/1bq6iytRnAzEnG9S\/o%3D"},{"@type":"BreadcrumbList","@id":"https:\/\/www.asm15.fr\/index.php\/2024\/07\/27\/flute-un-noyau-cuda-concu-pour-des-multiplications-de-matrices-quantifiees-fusionnees-accelerant-linference-des-grands-modeles-de-langage\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Accueil","item":"https:\/\/www.asm15.fr\/"},{"@type":"ListItem","position":2,"name":"FLUTE : Un noyau CUDA con\u00e7u pour des multiplications de matrices quantifi\u00e9es fusionn\u00e9es, acc\u00e9l\u00e9rant l&rsquo;inf\u00e9rence des grands mod\u00e8les de langage"}]},{"@type":"WebSite","@id":"https:\/\/www.asm15.fr\/#website","url":"https:\/\/www.asm15.fr\/","name":"IA & TECHNOLOGIES","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.asm15.fr\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-FR"},{"@type":"Person","@id":"https:\/\/www.asm15.fr\/#\/schema\/person\/9e2a0039267955de05a7ce4282a905e6","name":"m","image":{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/www.asm15.fr\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/6fe5cecb08501f51d81efa38bdd65daa775b9224f6dc9617c8615be766310093?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/6fe5cecb08501f51d81efa38bdd65daa775b9224f6dc9617c8615be766310093?s=96&d=mm&r=g","caption":"m"},"sameAs":["https:\/\/www.asm15.fr"],"url":"https:\/\/www.asm15.fr\/index.php\/author\/m\/"}]}},"_links":{"self":[{"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/posts\/201","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/comments?post=201"}],"version-history":[{"count":0,"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/posts\/201\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/media?parent=201"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/categories?post=201"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.asm15.fr\/index.php\/wp-json\/wp\/v2\/tags?post=201"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}