Update README.md
Browse files
README.md
CHANGED
|
@@ -97,23 +97,10 @@ lets users enter Arabic or English text and inspect:
|
|
| 97 |
- original text spans and offsets;
|
| 98 |
- decoded text for round-trip verification.
|
| 99 |
|
| 100 |
-
## Training mixture
|
| 101 |
-
|
| 102 |
-
The tokenizer was trained on sampled Arabic and English text. The processed
|
| 103 |
-
training text is not distributed in this repository.
|
| 104 |
-
|
| 105 |
-
| Source bucket | Characters used |
|
| 106 |
-
| --- | ---: |
|
| 107 |
-
| Saudi schoolbooks | 29,856,611 |
|
| 108 |
-
| AraMix Arabic | 200,000,000 |
|
| 109 |
-
| FineWeb2 Arabic | 200,000,000 |
|
| 110 |
-
| FineWeb English | 200,000,000 |
|
| 111 |
-
| **Total** | **629,856,611** |
|
| 112 |
-
|
| 113 |
## Validation
|
| 114 |
|
| 115 |
The released validation run reported zero round-trip failures and zero unknown
|
| 116 |
-
tokens across samples from
|
| 117 |
and FineWeb English. Additional stress tests cover Modern Standard Arabic,
|
| 118 |
dialects, Arabizi, English, code, mathematics, and Unicode edge cases.
|
| 119 |
|
|
|
|
| 97 |
- original text spans and offsets;
|
| 98 |
- decoded text for round-trip verification.
|
| 99 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
## Validation
|
| 101 |
|
| 102 |
The released validation run reported zero round-trip failures and zero unknown
|
| 103 |
+
tokens across samples from schoolbooks, AraMix Arabic, FineWeb2 Arabic,
|
| 104 |
and FineWeb English. Additional stress tests cover Modern Standard Arabic,
|
| 105 |
dialects, Arabizi, English, code, mathematics, and Unicode edge cases.
|
| 106 |
|