Upload BPE tokenizer trained on WikiText-2
Browse files
README.md
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
---
|
| 3 |
+
license: apache-2.0
|
| 4 |
+
tags:
|
| 5 |
+
- tokenizer
|
| 6 |
+
- bpe
|
| 7 |
+
- wikitext2
|
| 8 |
+
- nlp
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# WikiText-2 BPE Tokenizer
|
| 12 |
+
|
| 13 |
+
A Byte Pair Encoding (BPE) tokenizer trained on the WikiText-2 dataset.
|
| 14 |
+
|
| 15 |
+
## Model Details
|
| 16 |
+
- **Vocabulary Size**: 30,000 tokens
|
| 17 |
+
- **Training Data**: WikiText-2 (Salesforce/wikitext)
|
| 18 |
+
- **Special Tokens**: [PAD], [UNK], [CLS], [SEP], [MASK]
|
| 19 |
+
- **Compression Ratio**: ~6.4 characters per token
|
| 20 |
+
|
| 21 |
+
## Usage
|
| 22 |
+
```python
|
| 23 |
+
from transformers import AutoTokenizer
|
| 24 |
+
|
| 25 |
+
tokenizer = AutoTokenizer.from_pretrained("Rogarcia18/wikitext2-bpe-tokenizer")
|
| 26 |
+
```
|
| 27 |
+
|
| 28 |
+
## Training Details
|
| 29 |
+
- Dataset: WikiText-2 (wikitext-2-v1)
|
| 30 |
+
- Preprocessing: Deduplication, <unk> removal, whitespace normalization, remove samples cases with less than 10 characters
|
| 31 |
+
- Architecture: BPE with HuggingFace tokenizers library
|