Commit
•
db497b1
1
Parent(s):
8e76605
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
---
|
| 2 |
license: odc-by
|
| 3 |
---
|
| 4 |
-
# TxT360:
|
| 5 |
<center><img src="llm360_logo(1).png" alt="k2 eval table" /></center>
|
| 6 |
|
| 7 |
## 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.
|
|
|
|
| 1 |
---
|
| 2 |
license: odc-by
|
| 3 |
---
|
| 4 |
+
# TxT360: A Top-Quality LLM Pre-training Dataset Requires the Perfect Blend
|
| 5 |
<center><img src="llm360_logo(1).png" alt="k2 eval table" /></center>
|
| 6 |
|
| 7 |
## 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.
|