Dataset · cardiffnlp
Tweet classification benchmark
cardiffnlp/tweet_eval
Seven tweet classification tasks covering irony, hate, offensive, stance, emoji, emotion, and sentiment.
What's inside
An English tweet dataset structured as a unified benchmark of seven multi-class classification tasks. All tasks share a consistent format and fixed train, validation, and test splits. The size ranges from 100k to 1M examples.
What you can do with it
- Evaluate tweet classification models
- Fine-tune BERT and RoBERTa for sentiment analysis
- Compare models on a unified benchmark
- Tasks
- Text classification
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 33.3K
- Likes
- 165
Models trained on it
- mradermacher/elythros-decider-1.7b-GGUFZero shot classification 698
- littlemex/prismyra-decision-35b-a3b-fp8Text classification 556
- WhyZosa/llm-course-hw3-doraText generation 486
- mradermacher/NanoTitan-NLI-GGUFText classification 369
- littlemex/prismyra-decision-qwen3.6-35b-a3b-fp8-36lText classification 319
- WhyZosa/llm-course-hw3-loraText generation 313
- littlemex/prismyra-decision-qwen3.6-35b-a3b-nvfp4-36lText classification 244
- ketimiroshnik/llm-course-hw3-doraText generation 220
How to load it
from datasets import load_dataset
ds = load_dataset("cardiffnlp/tweet_eval")