Update README.md
Browse files
README.md
CHANGED
@@ -28,3 +28,6 @@ which is slower to iterate over.
|
|
28 |
The logic only splits on spaces, so the chunks are likely to be slightly larger than 820 chars.
|
29 |
The dataset has been normalized into lower case, with accents and non-english characters removed.
|
30 |
Items with less than 200 chars or more than 1000 chars have been removed.
|
|
|
|
|
|
|
|
28 |
The logic only splits on spaces, so the chunks are likely to be slightly larger than 820 chars.
|
29 |
The dataset has been normalized into lower case, with accents and non-english characters removed.
|
30 |
Items with less than 200 chars or more than 1000 chars have been removed.
|
31 |
+
|
32 |
+
This dataset is processed for convenience, at the expense of losing some percentage of the tokens due to truncation,
|
33 |
+
(assuming the training minibatches are truncated to 128 tokens).
|