SlimPajama — LLM Pretraining Dataset
A heavily deduplicated and cleaned version of RedPajama. It removes low-quality and repetitive data to improve training efficiency.
Dataset Details
| Provider | Cerebras |
| Category | Pretraining |
| Size | 627B Tokens |
| License | Apache 2.0 |
| Downloads | 150k |
| Tags | English, Deduplicated, CommonCrawl |
from datasets import load_dataset
ds = load_dataset("cerebras/SlimPajama-627B")
Related datasets
- FineWeb — Pretraining or continued pretraining on high-quality English web text
- The Stack v2 — Pretraining code models across 600+ programming languages
- Cosmopedia — Synthetic-textbook pretraining for small models (the SmolLM recipe)
- FineWeb 2 — Pretraining or continued pretraining in languages other than English
Frequently asked questions
Can I use SlimPajama commercially?
Yes — SlimPajama is released under Apache 2.0, a permissive license that allows commercial use, including training models you ship in a product. Check the dataset card for attribution requirements before release.
How much data does SlimPajama contain, and do I need all of it?
SlimPajama contains 627B Tokens. You rarely need all of it: for style and format fine-tuning, a few hundred to a few thousand examples are enough — load a slice (e.g. split="train[:1000]") and scale up only if quality plateaus.
What is SlimPajama best used for?
Efficient English pretraining on a heavily deduplicated corpus. It belongs to the Pretraining section of our dataset hub, where you'll find alternatives and complementary sets.
← All datasets | Fine-Tuning Guide