--- title: ​I Ran OpenAI’s Original GPT — 1 Locally: Here is What... description: A retrospective look at OpenAI's original GPT-1 model from 2018, exploring how a 117-million-parameter, 234MB model that once required a research team, eight... canonical: https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: ​I Ran OpenAI’s Original GPT — 1 Locally: Here is What Changed in 8 Years | daily.dev og:description: A retrospective look at OpenAI's original GPT-1 model from 2018, exploring how a 117-million-parameter, 234MB model that once required a research team, eight... og:url: https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip og:image: https://api.daily.dev/og/posts/PldEr0lip.png og:image:alt: ​I Ran OpenAI’s Original GPT — 1 Locally: Here is What Changed in 8 Years og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # ​I Ran OpenAI’s Original GPT — 1 Locally: Here is What Changed in 8 Years **[Medium](https://daily.dev/sources/medium_js)** · 4 min read · 2 upvotes · 0 comments ## Summary A retrospective look at OpenAI's original GPT-1 model from 2018, exploring how a 117-million-parameter, 234MB model that once required a research team, eight GPUs, and months of empirical experimentation to train can now run effortlessly on an 8GB laptop. The piece covers GPT-1's decoder-only Transformer architecture, its training on the 4.5GB BookCorpus dataset of 7,404 books, and the engineering challenges of 2018 such as limited GPU VRAM and the lack of established scaling templates, contrasting them with today's trivial inference requirements. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How big is OpenAI's original GPT-1 model file in terms of storage? GPT-1 occupies about 234 megabytes of storage in half-precision (FP16) format. It has 117 million parameters, with each parameter taking 2 bytes of memory, making the entire weight matrix small enough to fit into an 8GB laptop's memory dozens of times over, using less space than a modern web browser instance. _Curious how far model efficiency has come since GPT-1? daily.dev surfaces AI architecture deep dives for exactly that kind of comparison._ ### What dataset was GPT-1 trained on? GPT-1 was trained on the BookCorpus dataset, roughly 4.5 gigabytes of plain text drawn from 7,404 unique unpublished books. OpenAI chose long-form narrative prose over fragmented sources like Wikipedia so the Transformer layers would learn long-range semantic patterns, multi-sentence character arcs, and macro-level grammatical coherence. _Understanding training data choices like this helps developers evaluate model behavior on daily.dev._ ### Why was training GPT-1 so difficult in 2018 despite its small size today? Training GPT-1 in 2018 was hard because there were no established templates for scaling deep Transformers, so Ilya Sutskever's team had to determine hyperparameters through empirical experimentation alone. Additionally, NVIDIA V100 GPUs only had 16GB of VRAM each, so the team built custom software to split the network across 8 separate GPUs. _Tracking how training infrastructure evolved helps engineers reason about today's model-building tradeoffs, a thread daily.dev keeps visible._ ## Similar posts on daily.dev - [Writing an LLM from scratch, part 28 -- training a base model from scratch on an RTX 3090](https://daily.dev/posts/writing-an-llm-from-scratch-part-28----training-a-base-model-from-scratch-on-an-rtx-3090-qv0nonoeu) · Hacker News · 2 upvotes · 0 comments - [I ran 14b and 20b local AI models on my laptop for the same tasks, and result was surprising](https://daily.dev/posts/i-ran-14b-and-20b-local-ai-models-on-my-laptop-for-the-same-tasks-and-result-was-surprising-vqclmy07p) · XDA Developers · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#deep-learning](https://daily.dev/tags/deep-learning), [#openai](https://daily.dev/tags/openai) [View this post on daily.dev](https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"​I Ran OpenAI’s Original GPT — 1 Locally: Here is What Changed in 8 Years","url":"https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip"},"datePublished":"2026-08-24T15:18:53.626Z","dateModified":"2026-09-14T08:33:40.913Z","description":"A retrospective look at OpenAI's original GPT-1 model from 2018, exploring how a 117-million-parameter, 234MB model that once required a research team, eight...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2f8bf566c5737dc0e49781e06935b85e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2f8bf566c5737dc0e49781e06935b85e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,deep-learning,openai","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"​I Ran OpenAI’s Original GPT — 1 Locally: Here is What Changed in 8 Years"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/i-ran-openai-s-original-gpt-1-locally-here-is-what-changed-in-8-years-plder0lip#faq","mainEntity":[{"@type":"Question","name":"How big is OpenAI's original GPT-1 model file in terms of storage?","acceptedAnswer":{"@type":"Answer","text":"GPT-1 occupies about 234 megabytes of storage in half-precision (FP16) format. It has 117 million parameters, with each parameter taking 2 bytes of memory, making the entire weight matrix small enough to fit into an 8GB laptop's memory dozens of times over, using less space than a modern web browser instance. Curious how far model efficiency has come since GPT-1? daily.dev surfaces AI architecture deep dives for exactly that kind of comparison."}},{"@type":"Question","name":"What dataset was GPT-1 trained on?","acceptedAnswer":{"@type":"Answer","text":"GPT-1 was trained on the BookCorpus dataset, roughly 4.5 gigabytes of plain text drawn from 7,404 unique unpublished books. OpenAI chose long-form narrative prose over fragmented sources like Wikipedia so the Transformer layers would learn long-range semantic patterns, multi-sentence character arcs, and macro-level grammatical coherence. Understanding training data choices like this helps developers evaluate model behavior on daily.dev."}},{"@type":"Question","name":"Why was training GPT-1 so difficult in 2018 despite its small size today?","acceptedAnswer":{"@type":"Answer","text":"Training GPT-1 in 2018 was hard because there were no established templates for scaling deep Transformers, so Ilya Sutskever's team had to determine hyperparameters through empirical experimentation alone. Additionally, NVIDIA V100 GPUs only had 16GB of VRAM each, so the team built custom software to split the network across 8 separate GPUs. Tracking how training infrastructure evolved helps engineers reason about today's model-building tradeoffs, a thread daily.dev keeps visible."}}]} ```