{"id":1750,"date":"2026-07-31T03:00:31","date_gmt":"2026-07-31T03:00:31","guid":{"rendered":"https:\/\/convly.ai\/?p=1750"},"modified":"2026-08-01T06:45:53","modified_gmt":"2026-08-01T06:45:53","slug":"ollama-docker","status":"publish","type":"post","link":"https:\/\/convly.ai\/fr\/ollama-docker\/","title":{"rendered":"Guide Docker pour Ollama (2026) : ex\u00e9cuter Ollama dans un conteneur avec acc\u00e9l\u00e9ration GPU"},"content":{"rendered":"<p>En cours d\u2019ex\u00e9cution <strong>Ollama dans Docker<\/strong> constitue la m\u00e9thode la plus propre pour garantir la reproductibilit\u00e9 d\u2019un serveur local de mod\u00e8les : la m\u00eame commande fonctionne sur un ordinateur portable, un serveur domestique ou une machine distante \u00e9quip\u00e9e d\u2019un GPU, sans qu\u2019aucun composant ne s\u2019installe sur le syst\u00e8me h\u00f4te. La configuration est simple, mais deux d\u00e9tails \u2014 le passage du GPU au conteneur et la persistance des volumes \u2014 sont \u00e0 l\u2019origine de la majeure partie des pertes de temps.<\/p>\n<div style=\"background:#faf9ff;border:1px solid #e6e1f5;border-left:4px solid #6d28d9;border-radius:10px;padding:18px 22px;margin:28px 0;\">\n<p style=\"margin:0 0 10px;font-weight:700;color:#4c1d95;font-size:14px;letter-spacing:.5px;text-transform:uppercase;\">Quick answer<\/p>\n<p style=\"margin:0;\">T\u00e9l\u00e9chargez et lancez l\u2019image officielle avec un volume nomm\u00e9 afin que les mod\u00e8les survivent aux red\u00e9marrages : <code>docker run -d -v ollama:\/root\/.ollama -p 11434:11434 --name ollama ollama\/ollama<\/code>. Pour un GPU NVIDIA, installez le \u00ab NVIDIA Container Toolkit \u00bb sur la machine h\u00f4te, puis ajoutez l\u2019option <code>--gpus=all<\/code>. Ensuite, t\u00e9l\u00e9chargez les mod\u00e8les depuis l\u2019int\u00e9rieur du conteneur \u00e0 l\u2019aide de la commande <code>docker exec -it ollama ollama pull llama3.1<\/code>. L\u2019API est accessible sur le port 11434 exactement comme dans une installation native.<\/p>\n<\/div>\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_85 counter-flat ez-toc-counter ez-toc-container-direction\">\n<label for=\"ez-toc-cssicon-toggle-item-6a705a5a8b39f\" class=\"ez-toc-cssicon-toggle-label\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Basculer<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #000000;color:#000000\" 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: #000000;color:#000000\" 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><\/label><input type=\"checkbox\"  id=\"ez-toc-cssicon-toggle-item-6a705a5a8b39f\"  aria-label=\"Basculer\" \/><nav><ul class='ez-toc-list ez-toc-list-level-1' ><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/convly.ai\/fr\/ollama-docker\/#The_three_commands_that_matter\" >Les trois commandes essentielles<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/convly.ai\/fr\/ollama-docker\/#GPU_passthrough_the_part_that_fails\" >Passage du GPU au conteneur : l\u2019\u00e9tape qui \u00e9choue le plus souvent<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/convly.ai\/fr\/ollama-docker\/#Keep_your_models_between_restarts\" >Conserver vos mod\u00e8les entre les red\u00e9marrages<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/convly.ai\/fr\/ollama-docker\/#docker_compose_for_a_setup_you_keep\" >docker compose, pour une configuration durable<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/convly.ai\/fr\/ollama-docker\/#Frequently_asked_questions\" >Questions fr\u00e9quemment pos\u00e9es<\/a><\/li><\/ul><\/nav><\/div>\n<h2><span class=\"ez-toc-section\" id=\"The_three_commands_that_matter\"><\/span>Les trois commandes essentielles<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<div style=\"background:#faf9ff;border:1px solid #e6e1f5;border-radius:10px;padding:18px 22px;margin:28px 0;overflow-x:auto;\">\n<table style=\"width:100%;border-collapse:collapse;font-size:15px;\">\n<tr>\n<th style=\"text-align:left;padding:8px 10px;border-bottom:2px solid #e6e1f5;color:#4c1d95;\">Objectif<\/th>\n<th style=\"text-align:left;padding:8px 10px;border-bottom:2px solid #e6e1f5;color:#4c1d95;\">Commande<\/th>\n<\/tr>\n<tr>\n<td style=\"padding:8px 10px;border-bottom:1px solid #efecf8;\">Uniquement CPU<\/td>\n<td style=\"padding:8px 10px;border-bottom:1px solid #efecf8;\"><code>docker run -d -v ollama:\/root\/.ollama -p 11434:11434 --name ollama ollama\/ollama<\/code><\/td>\n<\/tr>\n<tr>\n<td style=\"padding:8px 10px;border-bottom:1px solid #efecf8;\">Avec GPU NVIDIA<\/td>\n<td style=\"padding:8px 10px;border-bottom:1px solid #efecf8;\"><code>docker run -d --gpus=all -v ollama:\/root\/.ollama -p 11434:11434 --name ollama ollama\/ollama<\/code><\/td>\n<\/tr>\n<tr>\n<td style=\"padding:8px 10px;border-bottom:1px solid #efecf8;\">T\u00e9l\u00e9charger un mod\u00e8le<\/td>\n<td style=\"padding:8px 10px;border-bottom:1px solid #efecf8;\"><code>docker exec -it ollama ollama pull llama3.1<\/code><\/td>\n<\/tr>\n<\/table>\n<\/div>\n<h2><span class=\"ez-toc-section\" id=\"GPU_passthrough_the_part_that_fails\"><\/span>Passage du GPU au conteneur : l\u2019\u00e9tape qui \u00e9choue le plus souvent<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Par d\u00e9faut, un conteneur ne peut pas acc\u00e9der \u00e0 votre GPU. Sous Linux, installez le <strong>NVIDIA Container Toolkit<\/strong> sur la machine h\u00f4te, red\u00e9marrez le service Docker, puis ajoutez l\u2019option <code>--gpus=all<\/code>. V\u00e9rifiez que la configuration fonctionne depuis l\u2019int\u00e9rieur du conteneur avec la commande <code>docker exec -it ollama nvidia-smi<\/code> \u2014 si cette commande \u00e9choue, Ollama s\u2019ex\u00e9cutera tout de m\u00eame, mais silencieusement sur le CPU ; vous conclurez alors \u00e0 tort que votre GPU est lent, alors qu\u2019il n\u2019est tout simplement pas utilis\u00e9. Sous Windows, utilisez Docker Desktop avec le backend WSL2 et des pilotes NVIDIA \u00e0 jour. Les utilisateurs AMD doivent employer l\u2019image marqu\u00e9e \u00ab ROCm \u00bb plut\u00f4t que l\u2019image par d\u00e9faut, tandis que les GPU Apple Silicon ne peuvent pas \u00eatre expos\u00e9s \u00e0 Docker du tout : sur Mac, ex\u00e9cutez Ollama nativement.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Keep_your_models_between_restarts\"><\/span>Conserver vos mod\u00e8les entre les red\u00e9marrages<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Les fichiers de mod\u00e8les sont volumineux et longs \u00e0 t\u00e9l\u00e9charger \u00e0 nouveau ; un conteneur sans volume les supprime d\u00e8s sa suppression. L\u2019option <code>-v ollama:\/root\/.ollama<\/code> pr\u00e9sente dans chacune des commandes ci-dessus cr\u00e9e un volume nomm\u00e9 qui persiste entre les red\u00e9marrages, les mises \u00e0 niveau et les changements d\u2019image. Si vous pr\u00e9f\u00e9rez stocker ces fichiers dans un emplacement accessible directement, montez plut\u00f4t un r\u00e9pertoire h\u00f4te : <code>-v \/srv\/ollama:\/root\/.ollama<\/code>.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"docker_compose_for_a_setup_you_keep\"><\/span>docker compose, pour une configuration durable<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Pour toute configuration durable, un fichier docker-compose est pr\u00e9f\u00e9rable \u00e0 une longue commande docker run : il centralise la r\u00e9servation du GPU, la gestion du volume et l\u2019exposition du port, red\u00e9marre automatiquement le service et permet d\u2019ajouter ult\u00e9rieurement une interface web. Le conteneur expose le m\u00eame point de terminaison compatible OpenAI sur <code>http:\/\/localhost:11434<\/code>, de sorte que les applications ne peuvent pas distinguer une installation conteneuris\u00e9e d\u2019une installation native.<\/p>\n<div style=\"background:#fffdf5;border:1px solid #f0e6c8;border-left:4px solid #b45309;border-radius:10px;padding:18px 22px;margin:28px 0;\">\n<p style=\"margin:0 0 10px;font-weight:700;color:#92400e;font-size:14px;letter-spacing:.5px;text-transform:uppercase;\">Convly&#8217;s take<\/p>\n<p style=\"margin:0;\">Conteneurisez Ollama lorsque la machine est un serveur, et \u00e9vitez Docker si elle est votre ordinateur portable. Sur un serveur domestique ou une machine distante \u00e9quip\u00e9e d\u2019un GPU, la combinaison volume + docker-compose est effectivement sup\u00e9rieure : reproductible, \u00e9volutif et facile \u00e0 d\u00e9placer. Sur une machine personnelle \u2014 notamment un Mac, o\u00f9 le passage du GPU \u00e0 Docker est impossible \u2014 Docker ajoute une surcouche qui nuit aux performances sans apporter de r\u00e9el avantage. Le probl\u00e8me le plus courant que nous observons est un conteneur tournant silencieusement sur le CPU parce que le \u00ab NVIDIA Container Toolkit \u00bb n\u2019a jamais \u00e9t\u00e9 install\u00e9 ; ex\u00e9cutez donc \u00ab nvidia-smi \u00bb \u00e0 l\u2019int\u00e9rieur du conteneur avant d\u2019effectuer tout test de performance.<\/p>\n<\/div>\n<h2><span class=\"ez-toc-section\" id=\"Frequently_asked_questions\"><\/span>Questions fr\u00e9quemment pos\u00e9es<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<h3>Ollama dans Docker prend-il en charge les GPU ?<\/h3>\n<p>Oui, avec NVIDIA via le \u00ab NVIDIA Container Toolkit \u00bb et l\u2019option <code>--gpus=all<\/code>, et avec AMD via l\u2019image ROCm. Les GPU Apple Silicon ne peuvent pas \u00eatre expos\u00e9s \u00e0 Docker : ex\u00e9cutez Ollama nativement sous macOS.<\/p>\n<h3>O\u00f9 sont stock\u00e9s les mod\u00e8les dans la configuration Docker ?<\/h3>\n<p>\u00c0 l\u2019int\u00e9rieur du conteneur, dans le r\u00e9pertoire <code>\/root\/.ollama<\/code>. Mappez ce chemin vers un volume nomm\u00e9 ou un r\u00e9pertoire h\u00f4te, sinon les mod\u00e8les disparaissent d\u00e8s la suppression du conteneur.<\/p>\n<h3>Ollama est-il plus lent sous Docker ?<\/h3>\n<p>De fa\u00e7on n\u00e9gligeable sous Linux, \u00e0 condition que le passage du GPU soit correctement configur\u00e9. Une lenteur per\u00e7ue signifie presque toujours que le conteneur tourne sur le CPU, car le GPU n\u2019a pas \u00e9t\u00e9 correctement expos\u00e9.<\/p>\n<h3>Puis-je ex\u00e9cuter plusieurs mod\u00e8les dans un seul conteneur ?<\/h3>\n<p>Oui \u2014 t\u00e9l\u00e9chargez-en autant que vous le souhaitez ; Ollama les charge \u00e0 la demande. La limite est la m\u00e9moire : seuls les mod\u00e8les effectivement charg\u00e9s consomment de la RAM ou de la VRAM.<\/p>\n<p>Nouveau sur cet outil ? Commencez par le <a href='\/fr\/how-to-install-ollama-2026\/'>Guide d\u2019installation d\u2019Ollama<\/a>, puis v\u00e9rifiez les besoins mat\u00e9riels dans les <a href='\/fr\/ollama-system-requirements-2026\/'>exigences syst\u00e8me d\u2019Ollama<\/a>, ou dimensionner un mod\u00e8le avec le <a href='\/fr\/llm-vram-calculator\/'>Calculateur de VRAM<\/a>.<\/p>","protected":false},"excerpt":{"rendered":"<p>Running Ollama in Docker keeps models and dependencies contained and makes the setup portable between machines. Here is the working configuration \u2014 including GPU passthrough and persistent storage.<\/p>","protected":false},"author":1,"featured_media":1785,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[5],"tags":[1048,756,256,259,1047],"class_list":["post-1750","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-tools","tag-docker","tag-gpu","tag-local-llm","tag-ollama","tag-self-hosting"],"_links":{"self":[{"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/posts\/1750","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/comments?post=1750"}],"version-history":[{"count":2,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/posts\/1750\/revisions"}],"predecessor-version":[{"id":1860,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/posts\/1750\/revisions\/1860"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/media\/1785"}],"wp:attachment":[{"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/media?parent=1750"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/categories?post=1750"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/convly.ai\/fr\/wp-json\/wp\/v2\/tags?post=1750"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}