CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

convert.py : BPE fixes (#2938)

Browse Source 
* convert.py: BPE fixes?

* Remove unnecessary conditional in addl token error handling
This commit is contained in:
Kerfuffle
2023-09-02 23:52:13 -06:00
committed by GitHub
parent 340af42f09
commit cff7b0bf07
1 changed files with 28 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
convert.py
View File

@@ -323,15 +323,27 @@ class BpeVocab:
self.bpe_tokenizer = json.loads(open(str(fname_tokenizer), encoding="utf-8").read())
added_tokens: dict[str, int]
if fname_added_tokens is not None:
# FIXME: Verify that added tokens here _cannot_ overlap with the main vocab.
added_tokens = json.load(open(fname_added_tokens, encoding="utf-8"))
else:
added_tokens = {}
# Fall back to trying to find the added tokens in tokenizer.json
tokenizer_json_file = fname_tokenizer.parent / 'tokenizer.json'
if not tokenizer_json_file.is_file():
added_tokens = {}
else:
tokenizer_json = json.load(open(tokenizer_json_file, encoding="utf-8"))
added_tokens = dict(
(item['content'], item['id'])
for item in tokenizer_json.get('added_tokens', [])
# Added tokens here can be duplicates of the main vocabulary.
if item['content'] not in self.bpe_tokenizer )
vocab_size: int = len(self.bpe_tokenizer)
expected_ids = list(range(vocab_size, vocab_size + len(added_tokens)))
actual_ids = sorted(added_tokens.values())
if expected_ids != actual_ids:
raise Exception(f"Expected added token IDs to be sequential and start at {len(added_tokens)}; got {actual_ids}")
expected_end_id = vocab_size + len(actual_ids) - 1
raise Exception(f"Expected the {len(actual_ids)} added token ID(s) to be sequential in the range {vocab_size} - {expected_end_id}; got {actual_ids}")
items = sorted(added_tokens.items(), key=lambda text_idx: text_idx[1])
self.added_tokens_list = [text for (text, idx) in items]
@@ -345,10 +357,22 @@ class BpeVocab:
from transformers.models.gpt2 import tokenization_gpt2 # type: ignore[import]
byte_encoder = tokenization_gpt2.bytes_to_unicode()
byte_decoder = {v: k for k, v in byte_encoder.items()}
score = 0.0
for i, item in enumerate(tokenizer):
text: bytes = item.encode("utf-8")
score: float = -i
yield text, score, gguf.TokenType.USER_DEFINED
# FIXME: These shouldn't be hardcoded, but it's probably better than the current behavior?
if i <= 258 and text.startswith(b'<') and text.endswith(b'>'):
if i == 0 and text == b'<unk>':
toktype = gguf.TokenType.UNKNOWN
elif i == 1 or i == 2:
toktype = gguf.TokenType.CONTROL
elif i >= 3 and text.startswith(b'<0x'):
toktype = gguf.TokenType.BYTE
else:
toktype = gguf.TokenType.NORMAL
else:
toktype = gguf.TokenType.NORMAL
yield text, score, toktype
def added_tokens(self) -> Iterable[tuple[bytes, float, gguf.TokenType]]:
for text in self.added_tokens_list: