Real multi-turn conversations scraped from ShareGPT.com — a site where users shared their best ChatGPT chats. Contains 52k real human-AI conversation trees. The dataset that made Vicuna and OpenChat possible by teaching models natural multi-turn dialogue.
Dataset Details
Provider
RyokoAI
Category
Instruction / SFT
Size
52k Conversations
License
CC-BY-NC 4.0
Downloads
1.1M
Tags
Multi-Turn, Real-World, ChatGPT, Conversational
from datasets import load_dataset
ds = load_dataset("RyokoAI/ShareGPT52K")
Fine-tune with this dataset
Estimated VRAM to fine-tune with QLoRA (4-bit base model + LoRA adapters), using conservative defaults:
WildChat-1M — Training on real user conversations; safety and robustness research
LMSYS-Chat-1M — Studying real-world usage patterns across 25 models (non-commercial)
Python-Edu — Continued pretraining for Python code understanding
OpenHermes 2.5 — The default general-purpose SFT mix for 7B-13B fine-tunes
Frequently asked questions
Can I use ShareGPT 52K commercially?
Not in a product — ShareGPT 52K is released under CC-BY-NC 4.0, which restricts use to research and other non-commercial purposes. For commercial fine-tuning, pick a permissively licensed dataset from the same category instead.
How much data does ShareGPT 52K contain, and do I need all of it?
ShareGPT 52K contains 52k Conversations. You rarely need all of it: for style and format fine-tuning, a few hundred to a few thousand examples are enough — load a slice (e.g. split="train[:1000]") and scale up only if quality plateaus.
What is ShareGPT 52K best used for?
Teaching natural multi-turn dialogue (non-commercial). It belongs to the Instruction / SFT section of our dataset hub, where you'll find alternatives and complementary sets.