{"id":37319,"date":"2025-06-06T09:02:43","date_gmt":"2025-06-06T09:02:43","guid":{"rendered":"https:\/\/mon-agent-ia.fr\/blog\/?p=37319"},"modified":"2025-06-06T09:02:44","modified_gmt":"2025-06-06T09:02:44","slug":"ibm-integrates-reasoning-into-its-llms-thanks-to-granite-3-2","status":"publish","type":"post","link":"https:\/\/mon-agent-ia.fr\/blog\/en\/ibm-integrates-reasoning-into-its-llms-thanks-to-granite-3-2\/","title":{"rendered":"IBM integrates reasoning into its LLMs thanks to Granite 3.2"},"content":{"rendered":"<p class=\"wp-block-paragraph\">IBM once again asserts itself in the artificial intelligence market with its new version of language models, Granite 3.2. In line with current trends, this innovation emphasizes conditional reasoning capabilities, which allow LLM performance to be adapted according to task complexity. In a technological landscape where speed of execution and depth of analysis are paramount, IBM intends to meet the growing needs of businesses. With significant advances in image recognition and predictive analytics, this new generation of models promises to transform natural language processing.<\/p>\n\n<h2 class=\"wp-block-heading\">IBM&rsquo;s Revolutionary Innovations in Granite 3.2<\/h2>\n\n<p class=\"wp-block-paragraph\">In 2025, IBM presented version 3.2 of Granite, demonstrating a true commitment to innovation. With an approach that reconciles performance and flexibility, the company aims to integrate reasoning into its LLMs in an efficient and accessible manner. This version is not limited to cosmetic additions but introduces memorable features. Conditional Reasoning: A Key Advance<\/p>\n\n<h3 class=\"wp-block-heading\">Conditional reasoning is at the heart of this development, allowing reasoning capabilities to be enabled or disabled depending on the nature of the query. Kyra, a developer at IBM, explains that for simple questions, such as \u00ab\u00a0What is the capital of France?\u00a0\u00bb, a fast answer is desirable. However, for more complex queries, such as \u00ab\u00a0Solve this engineering equation!\u00a0\u00bb, the model can be broken down into several analytical steps. This principle ensures a balance between speed and depth.<\/h3>\n\n<p class=\"wp-block-paragraph\">Here are some examples of tasks that would benefit from this flexibility:<\/p>\n\n<p class=\"wp-block-paragraph\">General knowledge questions<\/p>\n\n<ul class=\"wp-block-list\"><li>Mathematical calculations<\/li><li>Engineering problem solving<\/li><li>The particle filter: a bold method<\/li><\/ul>\n\n<h3 class=\"wp-block-heading\">IBM adopts an innovative method called \u00ab\u00a0particle filtering.\u00a0\u00bb This means that multiple reasoning processes are evaluated simultaneously, allowing the model to synthesize effective solutions. This technique, developed in collaboration with Red Hat, facilitates focusing on the analyses that yield the best results while maintaining a dynamic approach. IBM thus distinguishes itself from competitors, such as Deepseek, by integrating reasoning directly into the foundation model.<\/h3>\n\n<p class=\"wp-block-paragraph\">Image Recognition: Another Dimension of Granite 3.2<\/p>\n\n<h2 class=\"wp-block-heading\">Granite 3.2 also addresses the major challenge of managing scanned documents. This lightweight model, with its 2 billion parameters, is specifically designed for image and text recognition in various document types, making this technology essential for financial institutions that process large volumes of archives. By offering data extraction capabilities focused on document specificities, IBM enables more efficient information processing, both for text and for graphs, formulas, and tables. Document Types<\/h2>\n\n<p class=\"wp-block-paragraph\">Image Recognition Features<\/p>\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>Invoices<\/th>\n<th>Amount and Date Extraction<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Analysis Reports<\/td>\n<td>Graph and Table Extraction<\/td>\n<\/tr>\n<tr>\n<td>Administrative Forms<\/td>\n<td>Completed Field Recognition<\/td>\n<\/tr>\n<tr>\n<td>Towards Advanced Predictive Analytics<\/td>\n<td>In terms of predictive analytics, Granite 3.2 brings refined models based not only on traditional machine learning but also on innovative approaches. Jim, an analyst at IBM, discusses the TTM (tiny time mixer) models they developed to meet the specific and varied needs of businesses. These models, although compact (from 1 to 5 million parameters), now offer enormous customization possibilities, allowing event predictions with an appropriate length of context.<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n<h3 class=\"wp-block-heading\">The proposed context lengths vary, ranging from 512 to 52, to better meet the specific needs of daily or weekly financial forecasts.<\/h3>\n\n<p class=\"wp-block-paragraph\">Financial Predictions<\/p>\n\n<p class=\"wp-block-paragraph\">Maintenance Planning<\/p>\n\n<ul class=\"wp-block-list\"><li>Anomaly Detection<\/li><li>Practical Use Cases of Granite 3.2<\/li><li>By deploying these advanced models, IBM is enabling companies to leverage artificial intelligence in concrete ways. Industries from finance to logistics are already seeing a significant impact from optimized analytics. For example, a bank using this system can efficiently analyze thousands of documents with confidence, increasing its productivity and responsiveness.<\/li><\/ul>\n\n<h2 class=\"wp-block-heading\">The field demonstration of Granite 3.2 reveals how this technology can be a game-changer. After implementing Granite 3.2, Bank X successfully reduced its file processing time by 30%, which not only improved its efficiency but also allowed its employees to focus on higher-value tasks. Real-Time Data Analysis<\/h2>\n\n<p class=\"wp-block-paragraph\">Another area where Granite 3.2&rsquo;s intelligence shines is logistics optimization. By integrating predictive analytics, companies can not only anticipate raw material needs, but also forecast market fluctuations and adapt their production. This results in greater peace of mind for both suppliers and customers.<\/p>\n\n<p class=\"wp-block-paragraph\">Business Benefits<\/p>\n\n<h3 class=\"wp-block-heading\">Measurable Impacts<\/h3>\n\n<p class=\"wp-block-paragraph\">Optimized decision-making<\/p>\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>Reduced operating costs<\/th>\n<th>Improved customer satisfaction<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Increased retention rates<\/td>\n<td>Increased productivity<\/td>\n<\/tr>\n<tr>\n<td>Time savings in internal processes<\/td>\n<td>Future prospects with IBM Granite 3.2<\/td>\n<\/tr>\n<tr>\n<td>The advancements in Granite 3.2 raise the question: how can these models further transform the artificial intelligence landscape in the years to come? The answer lies largely in companies&rsquo; adaptability to new technologies. IBM is committed to remaining at the forefront of innovation and constantly developing its models to meet the future challenges of computing and analytics.<\/td>\n<td>In a world seeking ever more efficient solutions, IBM continues to demonstrate creativity and agility in its product development. As business requirements evolve, the right answer may lie in these new artificial intelligence tools. How could your company benefit from these developments?<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n<h2 class=\"wp-block-heading\"><\/h2>\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n","protected":false},"excerpt":{"rendered":"<p>IBM once again asserts itself in the artificial intelligence market with its new version of language models, Granite 3.2. In line with current trends, this innovation emphasizes conditional reasoning capabilities, which allow LLM performance to be adapted according to task complexity. In a technological landscape where speed of execution and depth of analysis are paramount, [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":37313,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1398],"tags":[66220,66223,8833,3452,66226],"class_list":["post-37319","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-ai-en","tag-granite-3-2-en","tag-ibm-en","tag-integration-en","tag-llm-en","tag-razonment-en"],"_links":{"self":[{"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/posts\/37319","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/comments?post=37319"}],"version-history":[{"count":1,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/posts\/37319\/revisions"}],"predecessor-version":[{"id":37320,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/posts\/37319\/revisions\/37320"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/media\/37313"}],"wp:attachment":[{"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/media?parent=37319"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/categories?post=37319"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mon-agent-ia.fr\/blog\/wp-json\/wp\/v2\/tags?post=37319"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}