zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-08-29 18:26:37 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Fix decoding manually added tokens

Browse Source
This commit is contained in:
Billel Mokeddem 2024-12-16 13:04:07 +04:00
parent fc0674d2d8
commit 8c67480050
1 changed files with 5 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
llms/mlx_lm/tokenizer_utils.py
View File

@ -159,6 +159,8 @@ class BPEStreamingDetokenizer(StreamingDetokenizer):
def __init__(self, tokenizer): def __init__(self, tokenizer):
self.tokenizer = tokenizer
self.clean_spaces = tokenizer.clean_up_tokenization_spaces self.clean_spaces = tokenizer.clean_up_tokenization_spaces
# Extract the tokens in a list from id to text # Extract the tokens in a list from id to text
@ -201,6 +203,9 @@ class BPEStreamingDetokenizer(StreamingDetokenizer):
"utf-8", "replace" "utf-8", "replace"
) )
if is_added: if is_added:
# We need to manually encode and decode the added tokens in case special characters
# used for `\n` / `\t` have been manually added in the added tokens
v = self.tokenizer.decode(self.tokenizer.encode(v))
text += v text += v
if not text.endswith("\ufffd"): if not text.endswith("\ufffd"):
self.text += self._maybe_trim_space(text) self.text += self._maybe_trim_space(text)