Dataset · microsoft
228k reasoning traces with memento
microsoft/OpenMementos
228,557 reasoning traces broken into blocks with compressed summaries for context management.
What's inside
OpenMementos-228K contains 228,557 reasoning traces annotated with blocks and compressed summaries, known as mementos. The data is derived from OpenThoughts-v3. The language is English, modality is text, and the size ranges from 100k to 1M.
What you can do with it
- Train LLMs or LoRA for long reasoning with context management
- Configure segmentation and compression of reasoning chains
- Evaluate models on long-form reasoning
- Tasks
- Text generation
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 732
- Likes
- 67
How to load it
from datasets import load_dataset
ds = load_dataset("microsoft/OpenMementos")