{"id":34647,"date":"2025-07-10T10:27:59","date_gmt":"2025-07-10T08:27:59","guid":{"rendered":"https:\/\/www.graviton.at\/letterswaplibrary\/creating-a-dataset-for-fine-tuning-a-code-generation-llm-in-the-data-science-domain\/"},"modified":"2025-07-10T10:27:59","modified_gmt":"2025-07-10T08:27:59","slug":"creating-a-dataset-for-fine-tuning-a-code-generation-llm-in-the-data-science-domain","status":"publish","type":"post","link":"https:\/\/www.graviton.at\/letterswaplibrary\/creating-a-dataset-for-fine-tuning-a-code-generation-llm-in-the-data-science-domain\/","title":{"rendered":"Creating A Dataset For Fine-Tuning A Code Generation LLM In The Data Science Domain"},"content":{"rendered":"<p><!-- SC_OFF --><\/p>\n<div class=\"md\">\n<p>I want to create a dataset using source code from GitHub to fine-tune a code generation LLM, specifically in the data science domain. Since I don&#8217;t have the budget to use LLMs to generate descriptions for the input, I&#8217;m designing a dataset where both the input and output are code (all crawled from GitHub).<\/p>\n<p>Is there a pipeline that can help me create input-output code pairs with consistent context (i.e., the input should provide enough context for the output) and focus on a specific domain?<\/p>\n<\/div>\n<p><!-- SC_ON -->   submitted by   <a href=\"https:\/\/www.reddit.com\/user\/Comfortable-Class905\"> \/u\/Comfortable-Class905 <\/a> <br \/> <span><a href=\"https:\/\/www.reddit.com\/r\/datasets\/comments\/1lp3cid\/creating_a_dataset_for_finetuning_a_code\/\">[link]<\/a><\/span>   <span><a href=\"https:\/\/www.reddit.com\/r\/datasets\/comments\/1lp3cid\/creating_a_dataset_for_finetuning_a_code\/\">[comments]<\/a><\/span><\/p><div class='watch-action'><div class='watch-position align-right'><div class='action-like'><a class='lbg-style1 like-34647 jlk' href='javascript:void(0)' data-task='like' data-post_id='34647' data-nonce='614a020375' rel='nofollow'><img class='wti-pixel' src='https:\/\/www.graviton.at\/letterswaplibrary\/wp-content\/plugins\/wti-like-post\/images\/pixel.gif' title='Like' \/><span class='lc-34647 lc'>0<\/span><\/a><\/div><\/div> <div class='status-34647 status align-right'><\/div><\/div><div class='wti-clear'><\/div>","protected":false},"excerpt":{"rendered":"<p>I want to create a dataset using source code from GitHub to fine-tune a code generation LLM,&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[85],"tags":[],"class_list":["post-34647","post","type-post","status-publish","format-standard","hentry","category-datatards","wpcat-85-id"],"_links":{"self":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts\/34647","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/comments?post=34647"}],"version-history":[{"count":0,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts\/34647\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/media?parent=34647"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/categories?post=34647"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/tags?post=34647"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}