update tokenization.py
Browse files- tokenization_qwen.py +44 -15
tokenization_qwen.py
CHANGED
@@ -27,11 +27,21 @@ IMEND = "<|im_end|>"
|
|
27 |
# regular texts, the surface forms of special tokens need to be
|
28 |
# as different as possible to minimize the impact
|
29 |
EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
|
36 |
|
37 |
def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
|
@@ -42,6 +52,7 @@ def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
|
|
42 |
for token, rank in (line.split() for line in contents.splitlines() if line)
|
43 |
}
|
44 |
|
|
|
45 |
class QWenTokenizer(PreTrainedTokenizer):
|
46 |
"""QWen tokenizer."""
|
47 |
|
@@ -51,20 +62,35 @@ class QWenTokenizer(PreTrainedTokenizer):
|
|
51 |
self,
|
52 |
vocab_file,
|
53 |
errors="replace",
|
|
|
54 |
**kwargs,
|
55 |
):
|
56 |
super().__init__(**kwargs)
|
57 |
|
58 |
-
|
|
|
|
|
59 |
|
60 |
-
self.mergeable_ranks = _load_tiktoken_bpe(vocab_file) # type:
|
61 |
self.special_tokens = {
|
62 |
token: index
|
63 |
-
for index, token in
|
64 |
-
SPECIAL_TOKENS, start=len(self.mergeable_ranks)
|
65 |
-
)
|
66 |
}
|
67 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
enc = tiktoken.Encoding(
|
69 |
"Qwen",
|
70 |
pat_str=PAT_STR,
|
@@ -89,7 +115,7 @@ class QWenTokenizer(PreTrainedTokenizer):
|
|
89 |
def __getstate__(self):
|
90 |
# for pickle lovers
|
91 |
state = self.__dict__.copy()
|
92 |
-
del state[
|
93 |
return state
|
94 |
|
95 |
def __setstate__(self, state):
|
@@ -103,7 +129,6 @@ class QWenTokenizer(PreTrainedTokenizer):
|
|
103 |
)
|
104 |
self.tokenizer = enc
|
105 |
|
106 |
-
|
107 |
def __len__(self) -> int:
|
108 |
return self.tokenizer.n_vocab
|
109 |
|
@@ -126,13 +151,17 @@ class QWenTokenizer(PreTrainedTokenizer):
|
|
126 |
ids.append(self.mergeable_ranks.get(token))
|
127 |
return ids
|
128 |
|
129 |
-
def _add_tokens(
|
|
|
|
|
|
|
|
|
130 |
if not special_tokens and new_tokens:
|
131 |
-
raise ValueError(
|
132 |
for token in new_tokens:
|
133 |
surface_form = token.content if isinstance(token, AddedToken) else token
|
134 |
if surface_form not in SPECIAL_TOKENS:
|
135 |
-
raise ValueError(
|
136 |
return 0
|
137 |
|
138 |
def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
|
|
|
27 |
# regular texts, the surface forms of special tokens need to be
|
28 |
# as different as possible to minimize the impact
|
29 |
EXTRAS = tuple((f"<|extra_{i}|>" for i in range(205)))
|
30 |
+
# changed to use actual index to avoid misconfiguration with vocabulary expansion
|
31 |
+
SPECIAL_START_ID = 151643
|
32 |
+
SPECIAL_TOKENS = tuple(
|
33 |
+
enumerate(
|
34 |
+
(
|
35 |
+
(
|
36 |
+
ENDOFTEXT,
|
37 |
+
IMSTART,
|
38 |
+
IMEND,
|
39 |
+
)
|
40 |
+
+ EXTRAS
|
41 |
+
),
|
42 |
+
start=SPECIAL_START_ID,
|
43 |
+
)
|
44 |
+
)
|
45 |
|
46 |
|
47 |
def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
|
|
|
52 |
for token, rank in (line.split() for line in contents.splitlines() if line)
|
53 |
}
|
54 |
|
55 |
+
|
56 |
class QWenTokenizer(PreTrainedTokenizer):
|
57 |
"""QWen tokenizer."""
|
58 |
|
|
|
62 |
self,
|
63 |
vocab_file,
|
64 |
errors="replace",
|
65 |
+
extra_vocab_file=None,
|
66 |
**kwargs,
|
67 |
):
|
68 |
super().__init__(**kwargs)
|
69 |
|
70 |
+
# how to handle errors in decoding UTF-8 byte sequences
|
71 |
+
# use ignore if you are in streaming inference
|
72 |
+
self.errors = errors
|
73 |
|
74 |
+
self.mergeable_ranks = _load_tiktoken_bpe(vocab_file) # type: Dict[bytes, int]
|
75 |
self.special_tokens = {
|
76 |
token: index
|
77 |
+
for index, token in SPECIAL_TOKENS
|
|
|
|
|
78 |
}
|
79 |
|
80 |
+
# try load extra vocab from file
|
81 |
+
if extra_vocab_file is not None:
|
82 |
+
used_ids = set(self.mergeable_ranks.values()) | set(self.special_tokens.values())
|
83 |
+
extra_mergeable_ranks = _load_tiktoken_bpe(extra_vocab_file)
|
84 |
+
for token, index in extra_mergeable_ranks.items():
|
85 |
+
if token in self.mergeable_ranks:
|
86 |
+
logger.info(f"extra token {token} exists, skipping")
|
87 |
+
continue
|
88 |
+
if index in used_ids:
|
89 |
+
logger.info(f'the index {index} for extra token {token} exists, skipping')
|
90 |
+
continue
|
91 |
+
self.mergeable_ranks[token] = index
|
92 |
+
# the index may be sparse after this, but don't worry tiktoken.Encoding will handle this
|
93 |
+
|
94 |
enc = tiktoken.Encoding(
|
95 |
"Qwen",
|
96 |
pat_str=PAT_STR,
|
|
|
115 |
def __getstate__(self):
|
116 |
# for pickle lovers
|
117 |
state = self.__dict__.copy()
|
118 |
+
del state["tokenizer"]
|
119 |
return state
|
120 |
|
121 |
def __setstate__(self, state):
|
|
|
129 |
)
|
130 |
self.tokenizer = enc
|
131 |
|
|
|
132 |
def __len__(self) -> int:
|
133 |
return self.tokenizer.n_vocab
|
134 |
|
|
|
151 |
ids.append(self.mergeable_ranks.get(token))
|
152 |
return ids
|
153 |
|
154 |
+
def _add_tokens(
|
155 |
+
self,
|
156 |
+
new_tokens: Union[List[str], List[AddedToken]],
|
157 |
+
special_tokens: bool = False,
|
158 |
+
) -> int:
|
159 |
if not special_tokens and new_tokens:
|
160 |
+
raise ValueError("Adding regular tokens is not supported")
|
161 |
for token in new_tokens:
|
162 |
surface_form = token.content if isinstance(token, AddedToken) else token
|
163 |
if surface_form not in SPECIAL_TOKENS:
|
164 |
+
raise ValueError("Adding unknown special tokens is not supported")
|
165 |
return 0
|
166 |
|
167 |
def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
|