MCPcopy Create free account
hub / github.com/AI-Hypercomputer/maxtext / _dump_chars_to_textfile

Function _dump_chars_to_textfile

src/MaxText/train_tokenizer.py:44–63  ·  view source on GitHub ↗

Write part of a TFDS sentence dataset to lines in a text file. Args: dataset: tf.dataset containing string-data. maxchars: int: approximate number of characters to save from dataset. data_keys: tuple[str]: what keys in dataset to dump from. Returns: name of temp file with dataset

(dataset: tf.data.Dataset, maxchars: int = int(1e7), data_keys=("text",))

Source from the content-addressed store, hash-verified

42
43
44def _dump_chars_to_textfile(dataset: tf.data.Dataset, maxchars: int = int(1e7), data_keys=("text",)) -> tuple[str, int]:
45 """Write part of a TFDS sentence dataset to lines in a text file.
46 Args:
47 dataset: tf.dataset containing string-data.
48 maxchars: int: approximate number of characters to save from dataset.
49 data_keys: tuple[str]: what keys in dataset to dump from.
50 Returns:
51 name of temp file with dataset bytes, exact number of characters dumped.
52 """
53 char_count = 0
54 ds_iter = dataset.as_numpy_iterator()
55 temp_dir = tempfile.gettempdir()
56 with tempfile.NamedTemporaryFile(delete=False, prefix=os.path.join(temp_dir, "ds_chars")) as outfp:
57 while char_count < maxchars:
58 example = next(ds_iter)
59 for k in data_keys:
60 line = example[k] + b"\n"
61 char_count += len(line)
62 outfp.write(line)
63 return outfp.name, char_count
64
65
66def _train_sentencepiece(

Callers 1

_train_sentencepieceFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected