Skip to content

Dataset · QuixiAI

Dolphin synthetic instructions in Orca style

QuixiAI/dolphin

About 4.5 million synthetic FLANv2 examples with GPT-4 and GPT-3.5 answers for instruction tuning.

What's inside

Dolphin is an English text dataset attempting to replicate Microsoft Orca results. It contains about 1 million FLANv2 examples with GPT-4 augmentations and about 3.5 million FLANv2 examples with GPT-3.5 augmentations. It uses the submix and system prompt distribution from the Orca paper, including all 75K CoT in FLAN-1m.

What you can do with it

  • Train an instructional LLM
  • Fine-tune LoRA on synthetic answers
  • Compare with Orca-like models
Tasks
Text generation
Languages
English
Size
1M–10M rows
Downloads this month
1.8K
Likes
447

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("QuixiAI/dolphin")