tokenizer

scripts/tokenizer_datasets.py

@@ -2,19 +2,38 @@ tokenizer_datasets = [
     #
     # multilingual
     #
+    # 193 MB, 1,141,967
+    *[
+        {'kind': 'base', 'path': 'xu-song/cc100-samples', 'name': name, 'split': 'train', 'format': lambda n: n['text']}
+        for name in [
+            'am', 'ar', 'as', 'az', 'be', 'bg', 'bn', 'bn_rom', 'br',
+            'bs', 'ca', 'cs', 'cy', 'da', 'de', 'el', 'en', 'eo', 'es',
+            'et', 'eu', 'fa', 'ff', 'fi', 'fr', 'fy', 'ga', 'gd', 'gl',
+            'gn', 'gu', 'ha', 'he', 'hi', 'hi_rom', 'hr', 'ht', 'hu',
+            'hy', 'id', 'ig', 'is', 'it', 'ja', 'jv', 'ka', 'kk', 'km',
+            'kn', 'ko', 'ku', 'ky', 'la', 'lg', 'li', 'ln', 'lo', 'lt',
+            'lv', 'mg', 'mk', 'ml', 'mn', 'mr', 'ms', 'my', 'my_zaw',
+            'ne', 'nl', 'no', 'ns', 'om', 'or', 'pa', 'pl', 'ps', 'pt',
+            'qu', 'rm', 'ro', 'ru', 'sa', 'si', 'sc', 'sd', 'sk', 'sl',
+            'so', 'sq', 'sr', 'ss', 'su', 'sv', 'sw', 'ta', 'ta_rom',
+            'te', 'te_rom', 'th', 'tl', 'tn', 'tr', 'ug', 'uk', 'ur',
+            'ur_rom', 'uz', 'vi', 'wo', 'xh', 'yi', 'yo',
+            'zh-Hans', 'zh-Hant', 'zu',
+        ]
+    ],
     # 3.8 GB, 19,454,996
     *[
-        {'path': 'sentence-transformers/parallel-sentences-wikimatrix', 'data_dir': 'all', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['non_english']}
+        {'kind': 'base', 'path': 'sentence-transformers/parallel-sentences-wikimatrix', 'data_dir': 'all', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['non_english']}
         for i in range(0, 100, 5)
     ],
     # 3.17 GB, 2,226,907
     *[
-        {'path': 'ontocord/fineweb-permissive-multilingual-2m', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['text']}
+        {'kind': 'base', 'path': 'ontocord/fineweb-permissive-multilingual-2m', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['text']}
         for i in range(0, 100, 5)
     ],
     # 1.64 GB, 1,001,000
     *[
-        {'path': 'distily/c4_multilingual_1M', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['text']}
+        {'kind': 'base', 'path': 'distily/c4_multilingual_1M', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['text']}
         for i in range(0, 100, 5)
     ],
 
@@ -23,11 +42,11 @@ tokenizer_datasets = [
     #
     # 1.44 GB, 63,357
     *[
-        {'path': 'neuralwork/arxiver', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['abstract']}
+        {'kind': 'base', 'path': 'neuralwork/arxiver', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['abstract']}
         for i in range(0, 100, 5)
     ],
     *[
-        {'path': 'neuralwork/arxiver', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['markdown']}
+        {'kind': 'base', 'path': 'neuralwork/arxiver', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['markdown']}
         for i in range(0, 100, 5)
     ],
 
@@ -36,7 +55,7 @@ tokenizer_datasets = [
     #
     # 7.81 GB, ~2,804,025
     *[
-        {'path': 'rombodawg/code_bagel_hermes-2.5', 'split': f'train[{i}%:{i + 5}%]', 'format': '{input} {output}'}
+        {'kind': 'base', 'path': 'rombodawg/code_bagel_hermes-2.5', 'split': f'train[{i}%:{i + 5}%]', 'format': '{input} {output}'}
         for i in range(0, 100, 5)
     ],
 
@@ -45,9 +64,9 @@ tokenizer_datasets = [
     #
     # 3.18 GB, 1,010,500 - paper says that extracted is 6GB
     *[
-        {'path': 'JeanKaddour/minipile', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['text']}
+        {'kind': 'base', 'path': 'JeanKaddour/minipile', 'split': f'train[{i}%:{i + 5}%]', 'format': lambda n: n['text']}
         for i in range(0, 100, 5)
     ],
-    {'path': 'JeanKaddour/minipile', 'split': 'validation', 'format': lambda n: n['text']},
-    {'path': 'JeanKaddour/minipile', 'split': 'test', 'format': lambda n: n['text']},
+    {'kind': 'base', 'path': 'JeanKaddour/minipile', 'split': 'validation', 'format': lambda n: n['text']},
+    {'kind': 'base', 'path': 'JeanKaddour/minipile', 'split': 'test', 'format': lambda n: n['text']},
 ]