Update README.md
Browse files
README.md
CHANGED
|
@@ -34,7 +34,7 @@ Processed distribution of Google's [C4](https://www.tensorflow.org/datasets/cata
|
|
| 34 |
|
| 35 |
Uses the text data from [`allenai/c4`](https://huggingface.co/datasets/allenai/c4).
|
| 36 |
|
| 37 |
-
`en` subset only.
|
| 38 |
|
| 39 |
T5 tokenizer was applied to the text.
|
| 40 |
Distributed as a ragged array.
|
|
@@ -49,6 +49,13 @@ Download size of all shards:
|
|
| 49 |
| Test | 299M | 8 | 44M | 179K |
|
| 50 |
| **Total** | **296G** | _N/A_ | _N/A_ | _N/A_ |
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
Download everything via:
|
| 53 |
|
| 54 |
```bash
|
|
|
|
| 34 |
|
| 35 |
Uses the text data from [`allenai/c4`](https://huggingface.co/datasets/allenai/c4).
|
| 36 |
|
| 37 |
+
Includes `en` subset only.
|
| 38 |
|
| 39 |
T5 tokenizer was applied to the text.
|
| 40 |
Distributed as a ragged array.
|
|
|
|
| 49 |
| Test | 299M | 8 | 44M | 179K |
|
| 50 |
| **Total** | **296G** | _N/A_ | _N/A_ | _N/A_ |
|
| 51 |
|
| 52 |
+
The data is uncompressed, in order to preserve support for random-seeking.
|
| 53 |
+
`.data.npy` would probably benefit from compression, because token sequences exhibit patterns.
|
| 54 |
+
|
| 55 |
+
Tokenization achieves a ~44% compression ratio.
|
| 56 |
+
Allen AI's original gzipped JSONL text data achieved a ~61% compression ratio.
|
| 57 |
+
So tokenized is about 13% bigger.
|
| 58 |
+
|
| 59 |
Download everything via:
|
| 60 |
|
| 61 |
```bash
|