vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

py : improve BPE tokenizer support (#5189)

Browse source
This commit is contained in:
Sang-Kil Park 2024-01-29 18:24:19 +09:00 • committed by GitHub
parent fbe7dfa53c
commit e76627bcce
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 4 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

3
convert.py
View file

@ -334,7 +334,10 @@ class Params:
class BpeVocab: class BpeVocab:
def __init__(self, fname_tokenizer: Path, fname_added_tokens: Path | None) -> None: def __init__(self, fname_tokenizer: Path, fname_added_tokens: Path | None) -> None:
self.bpe_tokenizer = json.loads(open(str(fname_tokenizer), encoding="utf-8").read()) self.bpe_tokenizer = json.loads(open(str(fname_tokenizer), encoding="utf-8").read())
try:
self.vocab = self.bpe_tokenizer["model"]["vocab"] self.vocab = self.bpe_tokenizer["model"]["vocab"]
except:
self.vocab = self.bpe_tokenizer
added_tokens: dict[str, int] added_tokens: dict[str, int]
if fname_added_tokens is not None: if fname_added_tokens is not None:
# FIXME: Verify that added tokens here _cannot_ overlap with the main vocab. # FIXME: Verify that added tokens here _cannot_ overlap with the main vocab.