{"id":1534,"date":"2026-10-19T11:41:18","date_gmt":"2026-10-19T11:41:18","guid":{"rendered":"https:\/\/www.interviewbit.com\/varsity\/blog\/?p=1534"},"modified":"2026-10-21T09:18:11","modified_gmt":"2026-10-21T09:18:11","slug":"toolformer","status":"publish","type":"post","link":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/","title":{"rendered":"Toolformer Explained: How LLMs Taught Themselves to Use Tools"},"content":{"rendered":"\n<div class=\"wp-block-group vlf-insights is-layout-flow wp-block-group-is-layout-flow\">\n<p class=\"vlf-insights-title wp-block-paragraph\">Key Insights<\/p>\n\n\n\n<ul class=\"wp-block-list vlf-insights-list\">\n\n<li>Toolformer is a language-model training approach introduced in a 2023 NeurIPS paper that teaches models when and how to call external APIs.<\/li>\n\n\n<li>Toolformer uses self-supervised learning to generate candidate API calls, execute them, and retain useful calls based on their effect on subsequent token-prediction loss.<\/li>\n\n\n<li>The original Toolformer paper examines five tool categories: a calculator, question-answering system, search engine, translation system and calendar.<\/li>\n\n\n<li>Toolformer differs from function calling, which enables structured function requests, and ReAct, which combines reasoning, actions and observations in an iterative workflow.<\/li>\n\n\n<li>Modern tool-augmented LLMs can use search, retrieval, calculators and APIs without necessarily following Toolformer&rsquo;s original training method.<\/li>\n\n\n<li>Tool access can improve calculations and factual retrieval, but incorrect tool selection, unreliable outputs and faulty interpretation can still cause errors.<\/li>\n\n<\/ul>\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Large language models (LLMs) can generate fluent text but may struggle with precise arithmetic, factual lookup and tasks that require external information. The Toolformer paper, published by Meta AI researchers in 2023, explored how language models could learn to address these limitations by using external tools through APIs. Instead of relying entirely on human-labelled examples, Toolformer uses self-supervised learning to identify when tool calls improve a model&rsquo;s predictions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This approach offers an important perspective on what tool use in LLMs means and how tool-augmented language models can go beyond text generation. This article explains how Toolformer works, its training pipeline, the tools used in the original research, and how its approach compares with function calling and ReAct agents.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What Is Toolformer: The Meta AI Research Approach<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer is a language-model training approach introduced in <em>Toolformer: Language Models Can Teach Themselves to Use Tools<\/em>, a paper published at NeurIPS 2023. It trains a model to decide which APIs to call, when to call them, what arguments to provide and how to incorporate the returned results into subsequent predictions.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">How Toolformer differs from a conventional LLM<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A conventional LLM generates text by predicting tokens based on learned patterns and the context available to it. Toolformer extends this process by learning to insert API calls when external information or computation can improve its predictions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">For example, instead of generating an arithmetic answer from its learned patterns alone, the model can call a calculator and use the result in its subsequent text.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The important distinction is that Toolformer is a specific self-supervised training method, not a generic term for every AI model that can use tools.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What Problem Does Toolformer Solve?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer addresses the gap between generating plausible language and reliably performing tasks that require calculation, factual retrieval or specialised operations.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1999\" height=\"667\" src=\"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-training-pipeline-showing-candidate-api-generatio.png\" alt=\"Toolformer training pipeline showing candidate API generation, loss-based filtering and fine-tuning.\" class=\"wp-image-1536\" srcset=\"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-training-pipeline-showing-candidate-api-generatio.png 1999w, https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-training-pipeline-showing-candidate-api-generatio-300x100.png 300w, https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-training-pipeline-showing-candidate-api-generatio-1024x342.png 1024w, https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-training-pipeline-showing-candidate-api-generatio-768x256.png 768w, https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-training-pipeline-showing-candidate-api-generatio-1536x513.png 1536w\" sizes=\"auto, (max-width: 1999px) 100vw, 1999px\" \/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">Why LLMs struggle with arithmetic, facts, and dates<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Next-token prediction does not guarantee accurate arithmetic or access to current information. A model might produce an incorrect numerical result, rely on outdated knowledge or generate an inaccurate factual answer.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">External tools for LLMs can provide capabilities that text generation alone may not reliably supply. A calculator can perform arithmetic, while a search engine can retrieve relevant information.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">However, tool access does not eliminate errors. A model can choose the wrong tool, provide incorrect arguments or misinterpret a tool&rsquo;s response. The challenge is therefore not simply connecting an API to an LLM; it is teaching the model when using that API is worthwhile.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">How Does Toolformer Work?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">How Toolformer works can be understood through three stages: generating candidate API calls, filtering them based on loss reduction, and fine-tuning the model on the retained examples.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Step 1 &mdash; Generate candidate API calls<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The process begins with a language model and a small number of demonstrations for each API. Using these examples, the model generates candidate tool calls at selected positions in text, including the arguments needed to execute them.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The calls and their outputs are represented in the training sequence. This allows the researchers to examine whether adding a tool response improves the model&rsquo;s ability to predict the text that follows.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This is the foundation of Toolformer self-supervised learning: the model generates candidate training examples rather than depending entirely on manually labelled tool-use decisions.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Step 2 &mdash; Execute calls and filter by loss reduction<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The candidate API calls are executed to obtain their results. The model then evaluates the subsequent text with and without the returned tool information.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The Toolformer loss-reduction filter retains calls when including the tool response improves the prediction of subsequent tokens according to the paper&rsquo;s criterion.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This step matters because not every available tool is useful for every sentence. A calculator call may improve a numerical prediction, while an unnecessary search call may add no meaningful benefit.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The filtering process helps identify tool calls that contribute to the training objective instead of treating every generated call as useful.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Step 3 &mdash; Fine-tune the model<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The retained examples become self-supervised training data. The model is fine-tuned to predict both ordinary text and useful tool calls, learning patterns associated with when and how to use external capabilities.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">During inference, the model can generate a tool call, the surrounding application executes it, and the result is made available for subsequent generation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">This pipeline distinguishes Toolformer from an approach that simply prompts an existing LLM to call an API.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Which Tools Does Toolformer Use?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The original Toolformer research examines five tool categories: a calculator, a question-answering system, a search engine, a translation system and a calendar. Each addresses a different limitation of language generation.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Tool<\/th><th>Purpose<\/th><th>Example use<\/th><\/tr><\/thead><tbody><tr><td>Calculator<\/td><td>Arithmetic<\/td><td>Computing a numerical result<\/td><\/tr><tr><td>Question-answering system<\/td><td>Factual retrieval<\/td><td>Answering a specific factual question<\/td><\/tr><tr><td>Search engine<\/td><td>Information retrieval<\/td><td>Looking up relevant information<\/td><\/tr><tr><td>Translation system<\/td><td>Language translation<\/td><td>Translating a phrase into English<\/td><\/tr><tr><td>Calendar<\/td><td>Date information<\/td><td>Checking the current date<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">How tool selection depends on the task<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Tool selection depends on the problem. A calculator tool in LLMs is useful for numerical operations, while search tools can help retrieve information that may not be present in the model&rsquo;s available context. Translation tools support language conversion.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These are the tools examined in the original paper, not a comprehensive list of capabilities available to modern LLM systems.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Toolformer vs Function Calling vs ReAct Agents<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer, function calling and ReAct address related aspects of LLM tool use, but they differ in how tool-use behaviour is established and executed.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Aspect<\/th><th>Toolformer<\/th><th>Function calling<\/th><th>ReAct agents<\/th><\/tr><\/thead><tbody><tr><td>Core idea<\/td><td>Learn useful tool calls through training<\/td><td>Generate structured calls to defined functions<\/td><td>Combine reasoning steps with actions and observations<\/td><\/tr><tr><td>Behaviour<\/td><td>Self-supervised examples and fine-tuning<\/td><td>Model uses available function definitions and schemas<\/td><td>Prompting or training supports iterative interaction<\/td><\/tr><tr><td>Execution<\/td><td>External API executes the call<\/td><td>Application executes the requested function<\/td><td>Agent framework executes actions and returns observations<\/td><\/tr><tr><td>Typical use<\/td><td>Research into learned tool-use behaviour<\/td><td>Structured API and application integration<\/td><td>Multi-step tasks requiring repeated tool interaction<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Function calling in LLMs allows an application to expose defined functions and process the model&rsquo;s structured requests. For example, the<a href=\"https:\/\/developers.openai.com\/api\/docs\/guides\/function-calling\"> <\/a><a href=\"https:\/\/developers.openai.com\/api\/docs\/guides\/function-calling?\">OpenAI API &mdash; Function calling<\/a> documentation describes how models can request external functionality and receive tool outputs through an application-managed workflow.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ReAct agents, introduced in a separate research paper, interleave reasoning traces with actions and observations to support iterative task-solving.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">When to use each approach<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Function calling is useful when an application needs structured, schema-defined requests. ReAct is suitable for workflows that require multiple action-and-observation steps. Toolformer is primarily a research contribution demonstrating how models can learn when tool use improves their predictions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These approaches are not mutually exclusive. A modern agent can use function calling within a ReAct-style interaction loop.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What Does Toolformer Mean for Tool-Augmented LLMs Today?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer helped demonstrate how models could learn tool-use behaviour from self-generated examples. Its significance lies in the training method, rather than being the architecture behind every current tool-using system.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">From learned tool use to modern LLM workflows<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Contemporary tool-augmented LLMs can combine language generation with search, retrieval, calculators, databases and application APIs. Their implementations may rely on fine-tuning, structured function calls or orchestration frameworks.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">These systems share the broader goal of extending an LLM beyond its internally generated text, but they do not necessarily use Toolformer&rsquo;s original training pipeline.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Benefits and limitations<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">External tools can support precise calculations, factual retrieval and specialised operations. However, the overall workflow remains vulnerable to incorrect tool selection, unreliable outputs and faulty interpretation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Execution costs, latency, permissions and validation also matter when deploying tool-using systems. Tool access should therefore be treated as one component of system design, not a universal solution to hallucinations or reliability problems.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Conclusion<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer demonstrated a self-supervised method for teaching language models when to use external APIs and how to incorporate their results into subsequent predictions. Its candidate-generation pipeline, loss-based filtering and fine-tuning process distinguish it from broader approaches such as function calling and ReAct. Although modern tool-augmented systems use different architectures and orchestration methods, Toolformer remains a useful research example of extending transformer-based language modelling with external capabilities.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">To explore these ideas further, the <strong><a href=\"https:\/\/www.interviewbit.com\/varsity\/iit-delhi\/generative-ai\">Generative AI programme by IIT Delhi<\/a><\/strong>, delivered through Varsity by InterviewBit, covers transformer foundations, RAG, Toolformer and modern agentic AI workflows. These topics help learners connect language-model fundamentals with the design of AI systems that use external tools.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Frequently Asked Questions<\/h2>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Who introduced Toolformer, and when was the paper published?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer was introduced by researchers including Timo Schick and colleagues in the paper <em>Toolformer: Language Models Can Teach Themselves to Use Tools<\/em>, published at NeurIPS 2023. The work demonstrated a self-supervised method for learning API use.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Is Toolformer an open-source LLM that I can use directly?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer is a published research approach, and third-party implementations exist. However, an implementation should not automatically be treated as the original model or an officially maintained Meta AI release.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Does Toolformer require large amounts of human-labelled tool-use data?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The original method uses a small number of demonstrations per API to generate candidate training examples. It then applies a self-supervised filtering process rather than requiring every tool-use decision to be manually labelled.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">How does Toolformer decide whether an API call is useful?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">It compares the model&rsquo;s prediction loss with and without the tool response. Candidate calls are retained when the response improves prediction of subsequent text under the paper&rsquo;s filtering criterion.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Can Toolformer use tools that were not included in its original training?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The original method trains around specified APIs and demonstrations. Supporting a new tool requires suitable integration and, depending on the implementation, additional examples or training; arbitrary new tools are not automatically supported.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Is Toolformer the same as retrieval-augmented generation (RAG)?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No. RAG retrieves relevant information to provide context for generation. Toolformer learns when to call APIs that can support different operations, including calculation, translation and retrieval.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Does Toolformer eliminate hallucinations in language models?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No. Tools can improve selected tasks, but a model may still make incorrect calls, receive unreliable results or misinterpret an output. Tool use requires validation and appropriate safeguards.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Can Toolformer perform multi-step tasks like an AI agent?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer learns tool-call behaviour within language generation. Multi-step agents commonly use explicit orchestration loops to select actions, process observations and decide what to do next. The approaches are related but distinct.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">What computing resources are required to reproduce the Toolformer research?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Reproduction requires a suitable language model, implementations of the relevant tools, data processing and fine-tuning resources. The exact requirements depend on the model size and implementation.<\/p>\n\n\n\n<h3 class=\"wp-block-heading has-medium-font-size\">Why is Toolformer important for GenAI engineers to study?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Toolformer provides a concrete example of connecting transformer-based language modelling with external APIs and using a self-supervised signal to learn when tool use improves predictions. It also helps clarify the distinction between training a model to use tools and building an application that executes tool calls.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Resources<\/h2>\n\n\n\n<ul class=\"wp-block-list\"><li><a href=\"https:\/\/arxiv.org\/abs\/2302.04761?\">Original Toolformer paper &mdash; arXiv<\/a><\/li><li><a href=\"https:\/\/ai.meta.com\/research\/publications\/toolformer-language-models-can-teach-themselves-to-use-tools\/?\">AI at Meta &mdash; Toolformer research publication<\/a><\/li><li><a href=\"https:\/\/developers.openai.com\/api\/docs\/guides\/function-calling\">OpenAI API &mdash; Function calling<\/a><\/li><li><a href=\"https:\/\/arxiv.org\/abs\/2210.03629?\">ReAct &mdash; original research paper<\/a><\/li><\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Key Insights Large language models (LLMs) can generate fluent text but may struggle with precise arithmetic, factual lookup and tasks that require external information. The Toolformer paper, published by Meta AI researchers in 2023, explored how language models could learn to address these limitations by using external tools through APIs. Instead of relying entirely on [&hellip;]<\/p>\n","protected":false},"author":7,"featured_media":1563,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[7],"tags":[],"class_list":["post-1534","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-generative-ai"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":8}},"acf":{"reviewed_by":""},"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.8 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Toolformer Explained: How LLMs Taught Themselves to Use Tools<\/title>\n<meta name=\"description\" content=\"Explore how Toolformer uses self-supervised learning and loss-based filtering to teach LLMs tool use, and compare its approach with function calling and ReAct agents.\" \/>\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.interviewbit.com\/varsity\/blog\/toolformer\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Toolformer Explained: How LLMs Taught Themselves to Use Tools\" \/>\n<meta property=\"og:description\" content=\"Explore how Toolformer uses self-supervised learning and loss-based filtering to teach LLMs tool use, and compare its approach with function calling and ReAct agents.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/\" \/>\n<meta property=\"og:site_name\" content=\"Varsity Blog\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-19T11:41:18+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-10-21T09:18:11+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-explained-.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1010\" \/>\n\t<meta property=\"og:image:height\" content=\"673\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Varsity on Behalf of CEP IIT Delhi\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Varsity on Behalf of CEP IIT Delhi\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"9 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/\"},\"author\":{\"name\":\"Varsity on Behalf of CEP IIT Delhi\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#\\\/schema\\\/person\\\/7b5db9a94eddee529cd35968692d9c29\"},\"headline\":\"Toolformer Explained: How LLMs Taught Themselves to Use Tools\",\"datePublished\":\"2026-10-19T11:41:18+00:00\",\"dateModified\":\"2026-10-21T09:18:11+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/\"},\"wordCount\":1950,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/toolformer-explained-.webp\",\"articleSection\":[\"Generative AI\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/\",\"url\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/\",\"name\":\"Toolformer Explained: How LLMs Taught Themselves to Use Tools\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/toolformer-explained-.webp\",\"datePublished\":\"2026-10-19T11:41:18+00:00\",\"dateModified\":\"2026-10-21T09:18:11+00:00\",\"description\":\"Explore how Toolformer uses self-supervised learning and loss-based filtering to teach LLMs tool use, and compare its approach with function calling and ReAct agents.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/toolformer-explained-.webp\",\"contentUrl\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/toolformer-explained-.webp\",\"width\":1010,\"height\":673,\"caption\":\"Toolformer-explained\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/toolformer\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Toolformer Explained: How LLMs Taught Themselves to Use Tools\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/\",\"name\":\"Varsity Blog\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#organization\",\"name\":\"Varsity Blog\",\"url\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/08\\\/varsity-logo.png\",\"contentUrl\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/08\\\/varsity-logo.png\",\"width\":275,\"height\":64,\"caption\":\"Varsity Blog\"},\"image\":{\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/company\\\/varsity-by-interviewbit\\\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/#\\\/schema\\\/person\\\/7b5db9a94eddee529cd35968692d9c29\",\"name\":\"Varsity on Behalf of CEP IIT Delhi\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/bdb12370d043ff980b2b8f1ccb67f5a1f0e333aaca46cc35358b1af8b1d98334?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/bdb12370d043ff980b2b8f1ccb67f5a1f0e333aaca46cc35358b1af8b1d98334?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/bdb12370d043ff980b2b8f1ccb67f5a1f0e333aaca46cc35358b1af8b1d98334?s=96&d=mm&r=g\",\"caption\":\"Varsity on Behalf of CEP IIT Delhi\"},\"description\":\"Varsity by InterviewBit, in collaboration with CEP IIT Delhi, creates industry-relevant learning programmes designed to help learners build practical, in-demand skills. Through this author profile, we publish articles that complement our courses covering curriculum-aligned topics, foundational concepts, emerging trends, and advanced insights. Our goal is to help learners deepen their understanding beyond the classroom and apply their knowledge confidently in real-world contexts.\",\"url\":\"https:\\\/\\\/www.interviewbit.com\\\/varsity\\\/blog\\\/author\\\/varsity-on-behalf-of-cep-iit-delhi\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Toolformer Explained: How LLMs Taught Themselves to Use Tools","description":"Explore how Toolformer uses self-supervised learning and loss-based filtering to teach LLMs tool use, and compare its approach with function calling and ReAct agents.","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.interviewbit.com\/varsity\/blog\/toolformer\/","og_locale":"en_US","og_type":"article","og_title":"Toolformer Explained: How LLMs Taught Themselves to Use Tools","og_description":"Explore how Toolformer uses self-supervised learning and loss-based filtering to teach LLMs tool use, and compare its approach with function calling and ReAct agents.","og_url":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/","og_site_name":"Varsity Blog","article_published_time":"2026-10-19T11:41:18+00:00","article_modified_time":"2026-10-21T09:18:11+00:00","og_image":[{"width":1010,"height":673,"url":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-explained-.webp","type":"image\/webp"}],"author":"Varsity on Behalf of CEP IIT Delhi","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Varsity on Behalf of CEP IIT Delhi","Est. reading time":"9 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#article","isPartOf":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/"},"author":{"name":"Varsity on Behalf of CEP IIT Delhi","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#\/schema\/person\/7b5db9a94eddee529cd35968692d9c29"},"headline":"Toolformer Explained: How LLMs Taught Themselves to Use Tools","datePublished":"2026-10-19T11:41:18+00:00","dateModified":"2026-10-21T09:18:11+00:00","mainEntityOfPage":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/"},"wordCount":1950,"commentCount":0,"publisher":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#organization"},"image":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#primaryimage"},"thumbnailUrl":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-explained-.webp","articleSection":["Generative AI"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/","url":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/","name":"Toolformer Explained: How LLMs Taught Themselves to Use Tools","isPartOf":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#primaryimage"},"image":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#primaryimage"},"thumbnailUrl":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-explained-.webp","datePublished":"2026-10-19T11:41:18+00:00","dateModified":"2026-10-21T09:18:11+00:00","description":"Explore how Toolformer uses self-supervised learning and loss-based filtering to teach LLMs tool use, and compare its approach with function calling and ReAct agents.","breadcrumb":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#primaryimage","url":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-explained-.webp","contentUrl":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/10\/toolformer-explained-.webp","width":1010,"height":673,"caption":"Toolformer-explained"},{"@type":"BreadcrumbList","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/toolformer\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.interviewbit.com\/varsity\/blog\/"},{"@type":"ListItem","position":2,"name":"Toolformer Explained: How LLMs Taught Themselves to Use Tools"}]},{"@type":"WebSite","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#website","url":"https:\/\/www.interviewbit.com\/varsity\/blog\/","name":"Varsity Blog","description":"","publisher":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.interviewbit.com\/varsity\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#organization","name":"Varsity Blog","url":"https:\/\/www.interviewbit.com\/varsity\/blog\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#\/schema\/logo\/image\/","url":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/08\/varsity-logo.png","contentUrl":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-content\/uploads\/2026\/08\/varsity-logo.png","width":275,"height":64,"caption":"Varsity Blog"},"image":{"@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.linkedin.com\/company\/varsity-by-interviewbit\/"]},{"@type":"Person","@id":"https:\/\/www.interviewbit.com\/varsity\/blog\/#\/schema\/person\/7b5db9a94eddee529cd35968692d9c29","name":"Varsity on Behalf of CEP IIT Delhi","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/bdb12370d043ff980b2b8f1ccb67f5a1f0e333aaca46cc35358b1af8b1d98334?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/bdb12370d043ff980b2b8f1ccb67f5a1f0e333aaca46cc35358b1af8b1d98334?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/bdb12370d043ff980b2b8f1ccb67f5a1f0e333aaca46cc35358b1af8b1d98334?s=96&d=mm&r=g","caption":"Varsity on Behalf of CEP IIT Delhi"},"description":"Varsity by InterviewBit, in collaboration with CEP IIT Delhi, creates industry-relevant learning programmes designed to help learners build practical, in-demand skills. Through this author profile, we publish articles that complement our courses covering curriculum-aligned topics, foundational concepts, emerging trends, and advanced insights. Our goal is to help learners deepen their understanding beyond the classroom and apply their knowledge confidently in real-world contexts.","url":"https:\/\/www.interviewbit.com\/varsity\/blog\/author\/varsity-on-behalf-of-cep-iit-delhi\/"}]}},"_links":{"self":[{"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/posts\/1534","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/comments?post=1534"}],"version-history":[{"count":2,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/posts\/1534\/revisions"}],"predecessor-version":[{"id":1537,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/posts\/1534\/revisions\/1537"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/media\/1563"}],"wp:attachment":[{"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/media?parent=1534"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/categories?post=1534"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.interviewbit.com\/varsity\/blog\/wp-json\/wp\/v2\/tags?post=1534"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}