Commit
•
4d0b85c
1
Parent(s):
7dbb89e
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,4 +1,6 @@
|
|
| 1 |
---
|
| 2 |
license: odc-by
|
| 3 |
---
|
| 4 |
-
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: odc-by
|
| 3 |
---
|
| 4 |
+
# TxT360: a globally deduplicated dataset for LLM pretraining
|
| 5 |
+
<center><img src="llm360_logo(1).png" alt="k2 eval table" /></center>
|
| 6 |
+
## We introduce TxT360 (Trillion eXtracted Text) the first dataset to globally deduplicate 99 CommonCrawl snapshots and 14 commonly used non-web data sources (e.g. FreeLaw, PG-19, etc.) providing pretraining teams with a recipe to easily adjust data weighting and train the most performant models.
|