Dataset · OpenAssistant
Multilingual OASST1 conversation corpus
OpenAssistant/oasst1
161K messages across 35 languages with quality ratings and conversation trees for training assistants.
What's inside
OASST1 is a human-generated corpus of assistant conversations containing 161,443 messages across 35 languages and 461,292 quality ratings. It includes over 10,000 fully annotated conversation trees crowdsourced worldwide. Metadata specifies tabular and text modalities.
What you can do with it
- Fine-tuning assistant models via SFT
- Training reward models for RLHF
- Evaluating assistant response quality
- Languages
- English, Spanish, Russian, German, Polish, Thai, Vietnamese, Swedish and 4 more
- Size
- 10K–100K rows
- Downloads this month
- 45.3K
- Likes
- 1.6K
Models trained on it
- sileod/deberta-v3-base-tasksource-nliZero shot classification 8K
- TheBloke/CodeLlama-13B-oasst-sft-v10-GGUF 6.8K
- mradermacher/archangel_sft-kto_llama30b-i1-GGUF 3.1K
- mradermacher/Aitana-7B-S-Instruct-i1-GGUFText generation 2.8K
- TheBloke/Llama2-70B-OASST-SFT-v10-GGUFText generation 2.7K
- ltg/normistral-7b-warm-instructText generation 2.4K
- davanstrien/chat_topicsText classification 2.4K
- TheBloke/Llama-2-70B-OASST-1-200-GGUF 2K
How to load it
from datasets import load_dataset
ds = load_dataset("OpenAssistant/oasst1")