Datasets:
Size:
10B<n<100B
License:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,11 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
size_categories:
|
| 4 |
+
- 10B<n<100B
|
| 5 |
+
---
|
| 6 |
+
`sample-10BT` version of the [FineWeb dataset](https://huggingface.co/datasets/HuggingFaceFW/fineweb) tokenized using the gpt2 tokenizer and split into 100M tokens binary shards.
|
| 7 |
+
|
| 8 |
+
A shard is simply a 1D stream of `np.uint16` numbers which are the tokenized samples from the dataset, stored contiguously.
|
| 9 |
+
Each sample from the dataset was prefixed with the `<|endoftext|>` special token before being tokenized.
|
| 10 |
+
|
| 11 |
+
There are 103 training shards (under `train/` dir) and 1 shard for validation (under `val/`).
|