{"id":6380,"date":"2025-01-17T11:31:27","date_gmt":"2025-01-17T10:31:27","guid":{"rendered":"https:\/\/www.editions-eni.fr\/blog\/?p=6380"},"modified":"2025-01-17T11:38:20","modified_gmt":"2025-01-17T10:38:20","slug":"retrieval-augmented-generation","status":"publish","type":"post","link":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/","title":{"rendered":"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d’IA g\u00e9n\u00e9ratives comme GPT ou LLama."},"content":{"rendered":"\n[et_pb_section fb_built=”1″ _builder_version=”4.16″ _module_preset=”default” locked=”off” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_row _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_column type=”4_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” hover_enabled=”0″ global_colors_info=”{}” theme_builder_area=”et_body_layout” sticky_enabled=”0″]

Les grands mod\u00e8les de langage (LLMs<\/em>), comme ChatGPT, sont \u00e9videmment limit\u00e9s par le choix des donn\u00e9es qui ont servi \u00e0 les entra\u00eener. Le RAG vise alors \u00e0 g\u00e9n\u00e9rer des r\u00e9ponses plus pr\u00e9cises et actualis\u00e9es que ces mod\u00e8les tr\u00e8s g\u00e9n\u00e9riques.<\/strong><\/p>\n

Courant 2025 para\u00eetra la quatri\u00e8me \u00e9dition de notre ouvrage “Data Scientist et langage R” qui va s’enrichir des derni\u00e8res techniques d’IA g\u00e9n\u00e9ratives et, parmi celles-ci, la plus utile \u00e0 savoir coder aujourd’hui : le RAG ou Retrieval-Augmented Generation. Revenons un peu sur cette technique\u2026<\/strong><\/p>\n

Par Eva et Henri Laude, sp\u00e9cialistes de l\u2019IA, de la Data science (entre autres) et auteurs aux Editions ENI (voir tous leurs ouvrages<\/a>)<\/em><\/p>[\/et_pb_text][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Introduit en 2020, le Retrieval-Augmented Generation est une technique hybride qui exploite des bases de donn\u00e9es externes en conjonction avec un mod\u00e8le g\u00e9n\u00e9ratif que l’on veut am\u00e9liorer ou sp\u00e9cialiser au regard d’un contexte particulier (un sujet pr\u00e9cis, les documents confidentiels d’une organisation …). Il s’agit de d’int\u00e9grer des donn\u00e9es nouvelles dans le mod\u00e8le d\u00e9j\u00e0 entra\u00een\u00e9 sans entacher sa coh\u00e9rence ou son efficacit\u00e9.<\/p>\n

Cette technique est peu co\u00fbteuse et \u00e0 la port\u00e9e de toutes les organisations<\/strong>, car elle ne mobilise pas les moyens informatiques monstrueux n\u00e9cessaires \u00e0 l’entra\u00eenement d’un mod\u00e8le g\u00e9n\u00e9ratif fondamental et polyvalent (foundation model<\/em>) comme les GPT, les LLama …<\/p>\n

Les principales \u00e9tapes d’un tel processus sont la r\u00e9cup\u00e9ration d\u2019informations pertinentes (retriever<\/em>), puis\u00a0 la g\u00e9n\u00e9ration augment\u00e9e par ces informations.<\/p>[\/et_pb_text][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_schema.webp” alt=”schema RAG” title_text=”schema RAG” align=”center” _builder_version=”4.27.0″ _module_preset=”default” width=”80%” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Limiter les biais de l\u2019IA<\/h2>\n

 <\/p>\n

Pour manipuler les informations, les donn\u00e9es d’un LLM sont pr\u00e9-trait\u00e9es et index\u00e9es sous forme de grands vecteurs encod\u00e9s dans une forme accessible permettant de capturer les relations s\u00e9mantiques entre les mots des textes (on parle d\u2019embedding<\/em>).<\/p>\n

Dans le cas du RAG, ce r\u00e9seau s\u00e9mantique appliqu\u00e9 \u00e0 la question que l’on se pose facilite le travail du retriever<\/em> qui identifie et classe les documents les plus pertinents en fonction de la requ\u00eate \u00e9mise par l’utilisateur, sans pour autant rester servilement “scotch\u00e9”\u00a0 \u00e0 une liste de mots cl\u00e9s. Tout ceci produit une r\u00e9ponse compl\u00e8te et coh\u00e9rente, tenant compte \u00e0 la fois du contexte de la requ\u00eate, de connaissances externes et\u00a0 des immenses comp\u00e9tences du\u00a0 foundation model<\/em>.<\/p>\n

Quand les documents retrouv\u00e9s ou impos\u00e9s au RAG sont pertinents, le LLM n’est alors plus enferm\u00e9 dans des donn\u00e9es statiques,<\/strong> ce qui permet des r\u00e9ponses plus en phase avec le monde r\u00e9el et limite le ph\u00e9nom\u00e8ne d’hallucination des mod\u00e8les en les emp\u00eachant de sortir du champ s\u00e9mantique que l’on d\u00e9sire explorer.<\/p>\n

L’utilit\u00e9 du RAG est d’autant plus grande qu’il est important pour l’utilisateur de mettre en \u0153uvre un dialogue de qualit\u00e9 pour \u00e9changer avec le mod\u00e8le LLM.<\/strong> Par exemple, dans le domaine de la sant\u00e9, un syst\u00e8me RAG acc\u00e9dera aux derni\u00e8res \u00e9tudes de cas et permettra \u00e0 un m\u00e9decin d’\u00e9tendre le champ de ses diagnostics sans parcourir toute la litt\u00e9rature m\u00e9dicale.<\/p>[\/et_pb_text][et_pb_code _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_code][et_pb_code _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_code][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_column type=”4_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Correctifs, sp\u00e9culatifs, fusionnels\u00a0: les RAG en pleine \u00e9volution<\/h2>\n

<\/strong><\/p>\n

Les processus RAG ne cessent de s’am\u00e9liorer,<\/strong> souvent en int\u00e9grant un m\u00e9canisme de validation permettant de\u00a0 corriger les r\u00e9ponses en comparant les informations g\u00e9n\u00e9r\u00e9es \u00e0 des sources fiables avant de les pr\u00e9senter : on parle alors de “RAG correctif<\/strong>“.<\/p>\n

Ces processus correctifs ne sont pas \u00e9vidents \u00e0 mettre en place, car il faut disposer de questions et de r\u00e9ponses issues du domaine qui permettront la mise en place d’un retrieval evaluator<\/em> qui qualifiera les r\u00e9ponses comme \u00e9tant correctes, incorrectes ou ambigu\u00ebs. Les couples “question versus r\u00e9ponse correcte”<\/em> permettront l’am\u00e9lioration du mod\u00e8le (knowledge refinement<\/em>) et les autres inciteront \u00e0 it\u00e9rer le processus de recherche d’informations compl\u00e9mentaires.<\/p>\n

Plus sophistiqu\u00e9s encore on trouve des RAG sp\u00e9culatifs<\/strong> qui produisent plusieurs r\u00e9ponses potentielles et s\u00e9lectionnent la plus pertinente via un m\u00e9canisme de r\u00e9troaction, particuli\u00e8rement utile dans des contextes ambigus.<\/p>\n

On trouve \u00e9galement des RAG fusionnels <\/strong>qui synth\u00e9tisent des informations issues de multiples sources afin de cr\u00e9er une r\u00e9ponse \u00e9quilibr\u00e9e, comme le ferait un groupe d’experts du domaine. Comme dans une discussion entre humains, ceux-ci \u00e9tendent ou transforment parfois le libell\u00e9 et le champ s\u00e9mantique de la question pos\u00e9e. Les contradictions sont r\u00e9solues par une introspection guid\u00e9e, typiquement une forme de cha\u00eene de raisonnement (CoT<\/em>, pour Chain of Thought<\/em>). Ce questionnement permet de mettre en \u00e9vidence diverses\u00a0 incoh\u00e9rences, ce qui favorise une meilleure inexplicabilit\u00e9 du r\u00e9sultat.\u00a0 Comme dans la\u00a0 ma\u00efeutique de Socrate, on s\u2019appuie donc ici sur une certaine autonomie cognitive de l\u2019apprenant pour \u00e9merger \u00e0 la connaissance.<\/p>\n

Notez qu’il ne faut pas confondre RAG et “fine tuning”<\/strong> : le RAG travaille sur l’int\u00e9gration de nouvelles donn\u00e9es, alors que les techniques de fine tuning visent \u00e0 am\u00e9liorer les param\u00e8tres du mod\u00e8le sur des grands ensembles de donn\u00e9es cibl\u00e9es, ce qui est plus co\u00fbteux. \u00c9videmment, de nombreux travaux alternent ou int\u00e8grent simultan\u00e9ment ces deux techniques, mais nous sugg\u00e9rons au d\u00e9veloppeurs IA en herbe de commencer par le RAG avant de s’attaquer au fine tuning.<\/p>\n

Vous l’aurez compris, le travail sur l’am\u00e9lioration des mod\u00e8les rel\u00e8ve tout aussi bien des sciences cognitives, de la philosophie de la connaissance, que de la programmation ou des math\u00e9matiques de l’intelligence artificielle. Pour autant, chacun peu bricoler dans son coin une nouvelle IA et participer \u00e0 peu de frais \u00e0 cette petite r\u00e9volution.<\/p>[\/et_pb_text][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Pour aller plus loin, ne manquez pas la sortie de notre livre \u00ab\u00a0Data Scientist et langage R\u00a0\u00bb (4e<\/sup> \u00e9dition) <\/strong>pr\u00e9vue en 2025 aux Editions ENI et sur la plateforme ENI elearning pour les professionnels (si vous \u00eates abonn\u00e9s aux newsletters et r\u00e9seaux sociaux des Editions ENI ou des n\u00f4tres, vous en serez inform\u00e9s\u00a0\ud83d\ude09).<\/p>\n

D\u2019ici l\u00e0, la 3e<\/sup> \u00e9dition reste compl\u00e8tement d\u2019actualit\u00e9 pour qui souhaite une formation compl\u00e8te sur les data sciences et l\u2019usage du langage R\u00a0: <\/p>[\/et_pb_text][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/livre_data_scientist3.webp” alt=”livre Data Scientist et langage R” title_text=”livre_data_scientist3″ url=”https:\/\/www.editions-eni.fr\/livre\/data-scientist-et-langage-r-autoformation-aux-bases-de-l-intelligence-artificielle-dans-l-univers-de-la-data-3e-edition-9782409030994″ url_new_window=”on” align=”center” _builder_version=”4.27.0″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_image][\/et_pb_column][\/et_pb_row][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.16″ _module_preset=”default” locked=”off” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_row _builder_version=”4.16″ _module_preset=”default” custom_padding=”||27px|||” locked=”off” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_column type=”4_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_text _builder_version=”4.16″ text_font=”||||||||” text_line_height=”1.8em” header_font=”||||||||” header_2_font=”Overpass|300|||||||” header_2_font_size=”35px” header_2_line_height=”1.5em” custom_margin=”||0px||false|false” header_2_font_size_tablet=”30px” header_2_font_size_phone=”20px” header_2_font_size_last_edited=”on|phone” border_color_all=”#1a0a38″ global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Pour aller plus loin<\/span><\/h2>[\/et_pb_text][et_pb_divider color=”#3BB6D5″ divider_position=”center” divider_weight=”3px” _builder_version=”4.16″ max_width=”90px” max_width_tablet=”13%” max_width_last_edited=”off|desktop” custom_margin=”5px||||false|false” custom_padding=”|||” locked=”off” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_divider][\/et_pb_column][\/et_pb_row][et_pb_row column_structure=”1_4,1_4,1_4,1_4″ _builder_version=”4.24.3″ _module_preset=”default” global_colors_info=”{}” custom_css_main_element_last_edited=”on|phone” custom_css_main_element_tablet=”display:flex;” custom_css_main_element_phone=”display:inherit;” theme_builder_area=”et_body_layout”][et_pb_column type=”1_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2024\/12\/business-intelligence-avec-excel-et-copilot-des-donnees-brutes-a-l-analyse-strategique-3e-edition-9782409046780_XL.webp” alt=”Scratch et Raspberry Pi Projets maker pour s’initier \u00e0 l’\u00e9lectronique et \u00e0 la robotique” title_text=”scratch-et-raspberry-pi-projets-maker-pour-s-initier-a-l-electronique-et-a-la-robotique-2e-edition-9782409027901_M” url=”https:\/\/www.editions-eni.fr\/livre\/business-intelligence-avec-excel-et-copilot-des-donnees-brutes-a-l-analyse-strategique-3e-edition-9782409046780″ url_new_window=”on” _builder_version=”4.27.0″ _module_preset=”default” width_tablet=”85%” width_phone=”” width_last_edited=”on|tablet” custom_margin=”||10px||false|false” custom_margin_tablet=”||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” custom_padding_tablet=”||||false|false” custom_padding_phone=”” custom_padding_last_edited=”on|tablet” border_radii_last_edited=”off|desktop” box_shadow_style=”preset3″ global_colors_info=”{}” transform_styles__hover_enabled=”on|hover” transform_scale__hover_enabled=”on|hover” transform_translate__hover_enabled=”on|desktop” transform_rotate__hover_enabled=”on|desktop” transform_skew__hover_enabled=”on|desktop” transform_origin__hover_enabled=”on|desktop” transform_scale__hover=”105%|105%” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.16″ _module_preset=”default” text_font=”||||||||” text_text_color=”#3BB6D5″ background_color=”RGBA(0,0,0,0)” custom_margin=”||5px||false|false” custom_margin_tablet=”” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” custom_padding=”||||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Livre<\/p>[\/et_pb_text][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” custom_padding_tablet=”|10px|||false|false” custom_padding_phone=”|0px|||false|false” custom_padding_last_edited=”on|phone” global_colors_info=”{}” theme_builder_area=”et_body_layout”]Business Intelligence avec Excel et Copilot<\/strong>\nDes donn\u00e9es brutes \u00e0 l’analyse strat\u00e9gique[\/et_pb_text][\/et_pb_column][et_pb_column type=”1_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/l-ia-et-la-generation-de-texte-comprendre-et-maitriser-chatgpt-gemini-perplexity-mistral-claude-9782409047848_XL.webp” alt=”Flutter D\u00e9veloppez vos applications mobiles multiplateformes avec Dart” title_text=”presentiel-web” url=”https:\/\/www.editions-eni.fr\/livre\/l-ia-et-la-generation-de-texte-comprendre-et-maitriser-chatgpt-gemini-perplexity-mistral-claude-9782409047848″ url_new_window=”on” _builder_version=”4.27.0″ _module_preset=”default” width=”100%” width_tablet=”85%” width_phone=”” width_last_edited=”on|tablet” custom_margin=”||10px||false|false” custom_margin_tablet=”||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” box_shadow_style=”preset3″ global_colors_info=”{}” transform_styles__hover_enabled=”on|hover” transform_scale__hover=”105%|105%” transform_scale__hover_enabled=”on|hover” transform_translate__hover_enabled=”on|desktop” transform_rotate__hover_enabled=”on|desktop” transform_skew__hover_enabled=”on|desktop” transform_origin__hover_enabled=”on|desktop” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” text_font=”||||||||” text_text_color=”#3BB6D5″ background_color=”RGBA(0,0,0,0)” custom_margin=”5px||5px||false|false” custom_margin_tablet=”” custom_margin_phone=”” custom_margin_last_edited=”on|phone” custom_padding=”||||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Livre<\/p>[\/et_pb_text][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” custom_padding_tablet=”|10px|||false|false” custom_padding_phone=”|0px|||false|false” custom_padding_last_edited=”on|phone” global_colors_info=”{}” theme_builder_area=”et_body_layout”]L\u2019IA et la g\u00e9n\u00e9ration de texte<\/strong>\nComprendre et ma\u00eetriser ChatGPT, Gemini, Perplexity, Mistral, Claude[\/et_pb_text][\/et_pb_column][et_pb_column type=”1_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/business-intelligence-avec-python-creez-vos-outils-bi-de-a-a-z-9782409048043_XL.webp” alt=”Flutter D\u00e9veloppez vos applications mobiles multiplateformes avec Dart” title_text=”flutter-developpez-vos-applications-mobiles-multiplateformes-avec-dart-9782409025273_M (1)” url=”https:\/\/www.editions-eni.fr\/livre\/business-intelligence-avec-python-creez-vos-outils-bi-de-a-a-z-9782409048043″ url_new_window=”on” src_tablet=”” src_phone=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2024\/01\/le-cloud-prive-avec-openstack-guide-pratique-pour-l-architecture-l-administration-et-l-implementation-9782409038693_XL.jpg” src_last_edited=”on|phone” _builder_version=”4.27.0″ _module_preset=”default” width_tablet=”85%” width_phone=”” width_last_edited=”on|tablet” custom_margin=”||10px||false|false” custom_margin_tablet=”||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” box_shadow_style=”preset3″ global_colors_info=”{}” transform_styles__hover_enabled=”on|hover” transform_scale__hover=”105%|105%” transform_scale__hover_enabled=”on|hover” transform_translate__hover_enabled=”on|desktop” transform_rotate__hover_enabled=”on|desktop” transform_skew__hover_enabled=”on|desktop” transform_origin__hover_enabled=”on|desktop” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.16″ _module_preset=”default” text_font=”||||||||” text_text_color=”#3BB6D5″ background_color=”RGBA(0,0,0,0)” custom_margin=”||5px||false|false” custom_margin_tablet=”” custom_margin_phone=”” custom_margin_last_edited=”on|phone” custom_padding=”||||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Livre<\/p>[\/et_pb_text][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” custom_padding_tablet=”|10px|||false|false” custom_padding_phone=”|0px|||false|false” custom_padding_last_edited=”on|phone” global_colors_info=”{}” theme_builder_area=”et_body_layout”]Business Intelligence avec Python<\/strong>\nCr\u00e9ez vos outils BI de A \u00e0 Z[\/et_pb_text][\/et_pb_column][et_pb_column type=”1_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/microsoft-copilot-optimisez-votre-productivite-en-collaborant-avec-l-ia-9782409046742_XL.webp” alt=”Flutter D\u00e9veloppez vos applications mobiles multiplateformes avec Dart” title_text=”elearning-graphisme” url=”https:\/\/www.editions-eni.fr\/livre\/microsoft-copilot-optimisez-votre-productivite-en-collaborant-avec-l-ia-9782409046742″ url_new_window=”on” _builder_version=”4.27.0″ _module_preset=”default” width_tablet=”85%” width_phone=”” width_last_edited=”on|tablet” custom_margin=”||10px||false|false” custom_margin_tablet=”||||false|false” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” box_shadow_style=”preset3″ global_colors_info=”{}” transform_styles__hover_enabled=”on|hover” transform_scale__hover=”105%|105%” transform_scale__hover_enabled=”on|hover” transform_translate__hover_enabled=”on|desktop” transform_rotate__hover_enabled=”on|desktop” transform_skew__hover_enabled=”on|desktop” transform_origin__hover_enabled=”on|desktop” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.24.3″ _module_preset=”default” text_font=”||||||||” text_text_color=”#3BB6D5″ background_color=”RGBA(0,0,0,0)” custom_margin=”||5px||false|false” custom_margin_tablet=”” custom_margin_phone=”” custom_margin_last_edited=”on|desktop” custom_padding=”||||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Livre<\/p>[\/et_pb_text][et_pb_text _builder_version=”4.27.0″ _module_preset=”default” custom_padding_tablet=”|10px|||false|false” custom_padding_phone=”|0px|||false|false” custom_padding_last_edited=”on|phone” global_colors_info=”{}” theme_builder_area=”et_body_layout”]Microsoft Copilot<\/strong>\nOptimisez votre productivit\u00e9 en collaborant avec l\u2019IA[\/et_pb_text][\/et_pb_column][\/et_pb_row][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.16″ _module_preset=”default” custom_margin=”0px||0px||false|false” custom_padding=”0px||50px||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_row _builder_version=”4.16″ _module_preset=”default” background_color=”#7ebec5″ custom_margin=”40px||||false|false” custom_padding=”20px||0px|20px|false|false” border_radii=”off|20px|20px||” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_column type=”4_4″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_text _builder_version=”4.16″ text_font=”||||||||” text_line_height=”1.8em” header_font=”||||||||” header_2_font=”Overpass|300|||||||” header_2_font_size=”35px” header_2_line_height=”1.5em” header_3_line_height=”1.8em” custom_margin=”0px||0px||false|false” custom_padding=”0px||0px||false|false” header_2_font_size_tablet=”30px” header_2_font_size_phone=”20px” header_2_font_size_last_edited=”on|phone” border_color_all=”#1a0a38″ global_colors_info=”{}” theme_builder_area=”et_body_layout”]

POUR LES ENTREPRISES<\/span><\/h3>[\/et_pb_text][et_pb_text _builder_version=”4.16″ text_font=”||||||||” text_text_color=”#000000″ text_line_height=”1em” header_font=”||||||||” header_2_font=”Overpass|300|||||||” header_2_font_size=”35px” header_2_line_height=”1.5em” header_3_line_height=”1.8em” custom_margin=”0px||20px||false|false” header_2_font_size_tablet=”30px” header_2_font_size_phone=”20px” header_2_font_size_last_edited=”on|phone” border_color_all=”#1a0a38″ global_colors_info=”{}” theme_builder_area=”et_body_layout”]

D\u00e9couvrez nos solutions de formation pour vos \u00e9quipes et apprenants :<\/span><\/h4>[\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row column_structure=”1_3,1_3,1_3″ _builder_version=”4.16″ _module_preset=”default” background_color=”#7ebec5″ custom_padding=”|30px|30px|30px|false|false” border_radii=”off|||20px|20px” border_color_all=”#3BB6D5″ global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_column type=”1_3″ _builder_version=”4.16″ _module_preset=”default” border_color_all=”#3BB6D5″ global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2021\/06\/reflechir.jpg” alt=”R\u00e9fl\u00e9chir en amont” title_text=”reflechir” _builder_version=”4.16″ _module_preset=”default” border_color_all=”#000000″ box_shadow_style=”preset1″ global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_image][\/et_pb_column][et_pb_column type=”1_3″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2021\/10\/iconeElearning.png” alt=”elearning” title_text=”iconeElearning” align=”center” _builder_version=”4.16″ _module_preset=”default” custom_margin=”10px||||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.16″ _module_preset=”default” header_4_text_color=”#FFFFFF” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

En e-learning avec
notre offre pour les professionnels<\/h4>[\/et_pb_text][et_pb_button button_url=”https:\/\/www.eni-elearning.com\/fr\/” url_new_window=”on” button_text=”En savoir plus” button_alignment=”center” _builder_version=”4.16″ _module_preset=”default” custom_button=”on” button_text_color=”#EEEEEE” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_button][\/et_pb_column][et_pb_column type=”1_3″ _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”][et_pb_image src=”https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2021\/10\/iconeFormateur.png” alt=”formateur” title_text=”iconeFormateur” align=”center” _builder_version=”4.16″ _module_preset=”default” custom_margin=”10px||||false|false” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_image][et_pb_text _builder_version=”4.16″ _module_preset=”default” global_colors_info=”{}” theme_builder_area=”et_body_layout”]

Avec un formateur, en pr\u00e9sentiel ou \u00e0 distance<\/span><\/h4>[\/et_pb_text][et_pb_button button_url=”https:\/\/www.eni-service.fr\/” url_new_window=”on” button_text=”En savoir plus” button_alignment=”center” _builder_version=”4.16″ _module_preset=”default” custom_button=”on” button_text_color=”#FFFFFF” global_colors_info=”{}” theme_builder_area=”et_body_layout”][\/et_pb_button][\/et_pb_column][\/et_pb_row][\/et_pb_section]\n","protected":false},"excerpt":{"rendered":"

Les grands mod\u00e8les de langage (LLMs), comme ChatGPT, sont \u00e9videmment limit\u00e9s par le choix des donn\u00e9es qui ont servi \u00e0 les entra\u00eener. Le RAG vise alors \u00e0 g\u00e9n\u00e9rer des r\u00e9ponses plus pr\u00e9cises et actualis\u00e9es que ces mod\u00e8les tr\u00e8s g\u00e9n\u00e9riques. Courant 2025 para\u00eetra la quatri\u00e8me \u00e9dition de notre ouvrage “Data Scientist et langage R” qui va […]<\/p>\n","protected":false},"author":8,"featured_media":6381,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_et_pb_use_builder":"on","_et_pb_old_content":"","_et_gb_content_width":"","inline_featured_image":false,"footnotes":""},"categories":[18,10],"tags":[],"genre":[33,37],"class_list":["post-6380","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-paroles-d-experts","category-societe-tendances","genre-parole-dexpert","genre-societe-tendances"],"acf":[],"yoast_head":"\nRetrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d'IA g\u00e9n\u00e9ratives comme GPT ou LLama. - Eni Blog<\/title>\n<meta name=\"description\" content=\"Adapter un mod\u00e8le d'IA g\u00e9n\u00e9rative gr\u00e2ce au Retrieval-Augmented Generation (RAG) pour des r\u00e9ponses plus pr\u00e9cises et contextualis\u00e9es.\" \/>\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.editions-eni.fr\/blog\/retrieval-augmented-generation\/\" \/>\n<meta property=\"og:locale\" content=\"fr_FR\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d'IA g\u00e9n\u00e9ratives comme GPT ou LLama. - Eni Blog\" \/>\n<meta property=\"og:description\" content=\"Adapter un mod\u00e8le d'IA g\u00e9n\u00e9rative gr\u00e2ce au Retrieval-Augmented Generation (RAG) pour des r\u00e9ponses plus pr\u00e9cises et contextualis\u00e9es.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/\" \/>\n<meta property=\"og:site_name\" content=\"Eni Blog\" \/>\n<meta property=\"article:published_time\" content=\"2025-01-17T10:31:27+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2025-01-17T10:38:20+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1792\" \/>\n\t<meta property=\"og:image:height\" content=\"635\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Denis Poillerat\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u00c9crit par\" \/>\n\t<meta name=\"twitter:data1\" content=\"Denis Poillerat\" \/>\n\t<meta name=\"twitter:label2\" content=\"Dur\u00e9e de lecture estim\u00e9e\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/\",\"url\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/\",\"name\":\"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d'IA g\u00e9n\u00e9ratives comme GPT ou LLama. - Eni Blog\",\"isPartOf\":{\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp\",\"datePublished\":\"2025-01-17T10:31:27+00:00\",\"dateModified\":\"2025-01-17T10:38:20+00:00\",\"author\":{\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/#\/schema\/person\/a6d10ff07aa69cb7c10371334bbe60f6\"},\"description\":\"Adapter un mod\u00e8le d'IA g\u00e9n\u00e9rative gr\u00e2ce au Retrieval-Augmented Generation (RAG) pour des r\u00e9ponses plus pr\u00e9cises et contextualis\u00e9es.\",\"breadcrumb\":{\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#breadcrumb\"},\"inLanguage\":\"fr-FR\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-FR\",\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#primaryimage\",\"url\":\"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp\",\"contentUrl\":\"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp\",\"width\":1792,\"height\":635,\"caption\":\"Illustration RAG\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/www.editions-eni.fr\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d’IA g\u00e9n\u00e9ratives comme GPT ou LLama.\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/#website\",\"url\":\"https:\/\/www.editions-eni.fr\/blog\/\",\"name\":\"Eni Blog\",\"description\":\"Blog des Editions ENI\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.editions-eni.fr\/blog\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-FR\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.editions-eni.fr\/blog\/#\/schema\/person\/a6d10ff07aa69cb7c10371334bbe60f6\",\"name\":\"Denis Poillerat\",\"url\":\"https:\/\/www.editions-eni.fr\/blog\/author\/dpoillerat\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d'IA g\u00e9n\u00e9ratives comme GPT ou LLama. - Eni Blog","description":"Adapter un mod\u00e8le d'IA g\u00e9n\u00e9rative gr\u00e2ce au Retrieval-Augmented Generation (RAG) pour des r\u00e9ponses plus pr\u00e9cises et contextualis\u00e9es.","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.editions-eni.fr\/blog\/retrieval-augmented-generation\/","og_locale":"fr_FR","og_type":"article","og_title":"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d'IA g\u00e9n\u00e9ratives comme GPT ou LLama. - Eni Blog","og_description":"Adapter un mod\u00e8le d'IA g\u00e9n\u00e9rative gr\u00e2ce au Retrieval-Augmented Generation (RAG) pour des r\u00e9ponses plus pr\u00e9cises et contextualis\u00e9es.","og_url":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/","og_site_name":"Eni Blog","article_published_time":"2025-01-17T10:31:27+00:00","article_modified_time":"2025-01-17T10:38:20+00:00","og_image":[{"width":1792,"height":635,"url":"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp","type":"image\/webp"}],"author":"Denis Poillerat","twitter_card":"summary_large_image","twitter_misc":{"\u00c9crit par":"Denis Poillerat","Dur\u00e9e de lecture estim\u00e9e":"8 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/","url":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/","name":"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d'IA g\u00e9n\u00e9ratives comme GPT ou LLama. - Eni Blog","isPartOf":{"@id":"https:\/\/www.editions-eni.fr\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#primaryimage"},"image":{"@id":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#primaryimage"},"thumbnailUrl":"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp","datePublished":"2025-01-17T10:31:27+00:00","dateModified":"2025-01-17T10:38:20+00:00","author":{"@id":"https:\/\/www.editions-eni.fr\/blog\/#\/schema\/person\/a6d10ff07aa69cb7c10371334bbe60f6"},"description":"Adapter un mod\u00e8le d'IA g\u00e9n\u00e9rative gr\u00e2ce au Retrieval-Augmented Generation (RAG) pour des r\u00e9ponses plus pr\u00e9cises et contextualis\u00e9es.","breadcrumb":{"@id":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#breadcrumb"},"inLanguage":"fr-FR","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/"]}]},{"@type":"ImageObject","inLanguage":"fr-FR","@id":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#primaryimage","url":"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp","contentUrl":"https:\/\/www.editions-eni.fr\/blog\/wp-content\/uploads\/2025\/01\/RAG_blog.webp","width":1792,"height":635,"caption":"Illustration RAG"},{"@type":"BreadcrumbList","@id":"https:\/\/www.editions-eni.fr\/blog\/retrieval-augmented-generation\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.editions-eni.fr\/blog\/"},{"@type":"ListItem","position":2,"name":"Retrieval-Augmented Generation: comment adapter soi-m\u00eame les mod\u00e8les d’IA g\u00e9n\u00e9ratives comme GPT ou LLama."}]},{"@type":"WebSite","@id":"https:\/\/www.editions-eni.fr\/blog\/#website","url":"https:\/\/www.editions-eni.fr\/blog\/","name":"Eni Blog","description":"Blog des Editions ENI","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.editions-eni.fr\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-FR"},{"@type":"Person","@id":"https:\/\/www.editions-eni.fr\/blog\/#\/schema\/person\/a6d10ff07aa69cb7c10371334bbe60f6","name":"Denis Poillerat","url":"https:\/\/www.editions-eni.fr\/blog\/author\/dpoillerat\/"}]}},"_links":{"self":[{"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/posts\/6380","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/comments?post=6380"}],"version-history":[{"count":4,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/posts\/6380\/revisions"}],"predecessor-version":[{"id":6392,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/posts\/6380\/revisions\/6392"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/media\/6381"}],"wp:attachment":[{"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/media?parent=6380"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/categories?post=6380"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/tags?post=6380"},{"taxonomy":"genre","embeddable":true,"href":"https:\/\/www.editions-eni.fr\/blog\/wp-json\/wp\/v2\/genre?post=6380"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}