Dataset · stanfordnlp
CoQA: Conversational question answering over texts
stanfordnlp/coqa
Conversational QA dataset: 127k questions with answers from 8k conversations across seven domains.
What's inside
The card specifies 127k question-answer pairs and 8k conversations based on text passages from seven different domains. Questions are structured as a conversation, answers are given in free text, and supporting evidence spans are highlighted in the passage. Language is English, modality is text.
What you can do with it
- Train and evaluate conversational question answering models
- SFT for LLMs on text-grounded question answering
- Evaluate the quality of evidence extraction from passages
- Tasks
- Question answering
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 5.6K
- Likes
- 87
Models trained on it
- parameterlab/apricot_binary_coqa_deberta-v3-base_for_vicuna-7b-v1.5Text classification 21
- parameterlab/apricot_clustering_coqa_deberta-v3-base_for_vicuna-7b-v1.5Text classification 20
- sgarbi/t5-qa-builderText generation 13
- parameterlab/apricot_binary_coqa_deberta-v3-base_for_gpt-3.5-turbo-0125Text classification 12
- parameterlab/apricot_clustering_coqa_deberta-v3-base_for_gpt-3.5-turbo-0125Text classification 11
How to load it
from datasets import load_dataset
ds = load_dataset("stanfordnlp/coqa")