Dataset · tencent
Benchmark for checking in-context learning
tencent/CL-bench
1,899 tasks where the model must learn new rules and procedures from context rather than pre-training.
What's inside
CL-bench is a benchmark for evaluating the ability of language models to learn from context. Tasks require mastering new domain knowledge, rule systems, complex procedures, and patterns from empirical data. A total of 1,899 tasks in English. The format in the card is truncated.
What you can do with it
- Evaluating in-context learning in LLMs
- Testing instruction-tuned models
- Comparing models on tasks with new rules from context
- Tasks
- Text generation
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 1K
- Likes
- 149
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("tencent/CL-bench")