Dataset · Open-Orca
Deduplicated instructions without RLHF
Open-Orca/SlimOrca-Dedup
363k unique instruction examples without RLHF, cleaned of duplicates for training generative models.
What's inside
This is a deduplicated and unfiltered subset of SlimOrca: 363k unique examples with RLHF examples removed. Duplicates were removed using minhash and Jaccard similarity. Demonstration models were trained on the full SlimOrca rather than this version.
What you can do with it
- SFT fine-tuning of instruction LLMs
- LoRA training for question-answering
- Testing text generation on instructions
- Tasks
- Text classification, Question answering, Text generation
- Size
- 100K–1M rows
- Downloads this month
- 1.7K
- Likes
- 96
Models trained on it
- Felladrin/Llama-160M-Chat-v1Text generation 23.9K
- TheBloke/una-cybertron-7B-v2-GGUF 2K
- TheBloke/DiscoLM-70B-GGUFText generation 2K
- mradermacher/Medichat-Llama3-8B-i1-GGUF 1.9K
- mradermacher/Ryeta-0-i1-GGUF 1.5K
- TheBloke/phi-2-orange-GGUF 779
- pints-ai/1.5-Pints-2K-v0.1Text generation 697
- nitsuai/llama-3-neural-chat-v1-8b-GGUFText generation 679
How to load it
from datasets import load_dataset
ds = load_dataset("Open-Orca/SlimOrca-Dedup")