llm-training-data 开源项目推荐（40 个）

ServiceNow /SyGra

SyGra - Graph-oriented Synthetic data generation Pipeline

Cre4T3Tiv3 /Cre4T3Tiv3

该项目在 GitHub 上没有填写描述

b2bemaillists /b2b-email-leads-ranking

An open-source collection of datasets, guides, and rankings for B2B email marketing and lead generation. Your go-to resource for sales prospecting strategies.

lexicanum-imperialis /Warhammer-Fantasy-Battles-6th-Definitive-edition

A complete BattleScribe and New Recruit data repository for Warhammer Fantasy Battles 6th Edition. Includes all official and experimental rules, FAQs, and community contributions.

emailmarketingdataset /Open-Email-Marketing-Dataset

Following is the Open Email Marketing Dataset; you can use it without any restrictions.

TheFamLee /ministryplatform-help-center

Markdown mirror of the MinistryPlatform Help Center (help.acst.com) for querying by AI coding assistants via Context7 MCP. Content © ACS Technologies, mirrored with permission.

obrienciaran /MedScreen_filter_POC

(EXPERIMENTAL) Evidence-grounded data filter that checks PubMed claims against scientific evidence to curate medical AI training data.

unmodeled-tyler /DoW-UFO-UAP-1

Machine-readable dataset for public Department of War / PURSUE UFO-UAP Release 01 records.

BlazeWild /Custom_LLM_DataGen_Template

🔧 Modular pipeline for generating high-quality, domain-specific datasets for LLM fine-tuning — from PDFs and web scraping to synthetic Q&A generation, quality filtering, and training-ready formatting.

shimo4228 /doctrine-corpus

A CC0 bilingual (English and Japanese) Q&A corpus that encodes the documented judgment of the author's practice lines (AKC, Contemplative Agent, AAP, Authorship Strategy) as material for LLM training and RAG; the operational form of Authorship Strategy's LLM-first ingest tactic.

Sbussiso /synthetic-self-correction-and-thinking-samples

This is a collection of synthetically generated datapoints for fine tuning meant for seeding larger synthetic datasets.

babyjack-svg /warmflow-content

暖流人心 心理健康內容庫｜臨床心理師劉子維｜AEO mirror of warmflowpsy.tw

deepakshroff /Capston-Gemini-ChatBot

👨‍🏫This project was developed under the guidance of Mr. Lokesh Sir as part of the AI & ML Training Program. It explores LLM integration using Google Gemini APIs with a custom UI built on Streamlit.

AmanPriyanshu /Stratified-LLM-Subsets-100K-1M-Scale

Stratified LLM Subsets delivers diverse training data at 100K-1M scales across pre-training (FineWeb-Edu, Proof-Pile-2), instruction-following (Tulu-3, Orca AgentInstruct), and reasoning distillation (Llama-Nemotron). Embedding-based k-means clustering ensures maximum diversity across 5 high-quality open datasets.

ChrisRoboMacro /Global-Macro-Economic-Simulations-and-Financial-Market-Responses

Free 10-simulation sample (CC BY 4.0) of RoboMacro's Global Macro Economic Simulations: 30 countries, 45 macro and market series, Q1-Q20 impulse responses as JSON training records, QA and tool-use rows. 1,000- and 10,000-sim packs for exclusive sale to one AI lab: [email protected]

jakejb53 /dataset-tools

Scripts for preparing LLM training datasets

dollce /getw

web page into a source-grounded semantic document and derives LLM-ready text from that document

vinsblack /The-Stack-Processed-v2

Sample edition of The Stack Enriched: annotated, secure, and optimized code dataset, this is a sample version

zhangyinruizhang-lgtm /bilibili-danmaku-scraper

Bilibili Danmaku Scraper API (哔哩哔哩弹幕提取器) — Extract real-time XML bullet comments (弹幕), exact video timestamps, peak reaction minutes (peakMinute), and sender hashes from Bilibili videos without cookies or login. Built for AI/LLM corpora, Python, n8n & Claude MCP.

Radiationpatterngordianknot284 /uap-pursue-release-01

Archive the first official U.S. Department of War UAP declassification, containing 162 verified technical files, photographs, and videos.