grokx

Appearance

English library

grokx

Grok’s Multi-Source Tuning Benchmark

Play icon crypto ? OpenAI deepseek grok

📊 Tulu 3 SFT Mixture

A 939k-sample dataset for training Tulu 3 models, blending diverse prompts from math, code, and dialogues under ODC-BY-1.0 license.

Explore Dataset

🧠 Dataset Composition

Includes 18 sources like CoCoNot (10k), FLAN v2 (89k), and Aya (100k), covering varied tasks for robust AI fine-tuning.

Learn More

✅ Licensing

Licensed under ODC-BY-1.0, with subsets like No Robots (CC-BY-NC-4.0) and Aya (Apache 2.0), some non-commercial.

Discover Process

✔️ Data Structure

Each sample has an ID, messages (user prompts + responses), and source, optimized for instruction-tuning.

Check Details

📚 Tulu 3 Models

Powers Llama-3.1-based Tulu 3 series (8B/70B) across SFT, DPO, and RLVR stages for advanced AI performance.

Try It Out

Find the plan that's right for you, each plan includes

docs iconsDocs
sheets iconsSheets
slides iconsslides
forms iconsforms
keep iconskeep
sites iconssites
drive iconsdrive
gmail iconsgmail
meet iconsmeet
calendar iconscalendar
Chat_icon@1x iconsChat
docusaurus_keytar iconsjup
docusaurus iconsBusiness
GoogleMaps iconsGoogleMaps
book iconbook
books iconbooks
security iconsecurity
restaurant iconrestaurant
thought iconthought
recipe iconrecipe
news iconnews
deepseek icondeepseek
deepseekr1 icondeepseekr1
deepseekr2 icondeepseekr2
deepseekr2 icondeepseekr3
deepseekr7 icondeepseekr7

Released under the MIT License.

Copyright © 2014-2025 grokx