Remove nested directory: BitTransformerLM/build_full_bits.py
Browse files
BitTransformerLM/build_full_bits.py
DELETED
|
@@ -1,23 +0,0 @@
|
|
| 1 |
-
import pathlib
|
| 2 |
-
import torch
|
| 3 |
-
from datasets import load_dataset
|
| 4 |
-
|
| 5 |
-
TXT_MB = 100
|
| 6 |
-
OUT = pathlib.Path('full_bits.pt')
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
def build_bits(out: pathlib.Path = OUT, txt_mb: int = TXT_MB) -> None:
|
| 10 |
-
ds = load_dataset('wikitext', 'wikitext-2-raw-v1', split='train')
|
| 11 |
-
buf = bytearray()
|
| 12 |
-
for line in ds['text']:
|
| 13 |
-
buf.extend(line.encode() + b"\n")
|
| 14 |
-
if len(buf) >= txt_mb * 2 ** 20:
|
| 15 |
-
break
|
| 16 |
-
bits = []
|
| 17 |
-
for byte in buf:
|
| 18 |
-
bits.extend(int(b) for b in f'{byte:08b}')
|
| 19 |
-
tensor = torch.tensor(bits, dtype=torch.uint8)
|
| 20 |
-
torch.save(tensor, out)
|
| 21 |
-
|
| 22 |
-
if __name__ == '__main__':
|
| 23 |
-
build_bits()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|