Spaces:
Running
Running
File size: 7,394 Bytes
909e36b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 |
"""Tokenization classes for vibevoice."""
from typing import List, Optional, Union
from transformers.utils import logging
from transformers.models.qwen2.tokenization_qwen2 import Qwen2Tokenizer
from transformers.models.qwen2.tokenization_qwen2_fast import Qwen2TokenizerFast
logger = logging.get_logger(__name__)
class VibeVoiceTextTokenizer(Qwen2Tokenizer):
"""
Construct a VibeVoice tokenizer. Based on the Qwen2 tokenizer with additional special tokens for speech.
Args:
vocab_file (`str`):
Path to the vocabulary file.
merges_file (`str`):
Path to the merges file.
errors (`str`, *optional*, defaults to `"replace"`):
Paradigm to follow when decoding bytes to UTF-8.
unk_token (`str`, *optional*, defaults to `"<|endoftext|>"`):
The unknown token.
bos_token (`str`, *optional*):
The beginning of sequence token. Not used for vibevoice.
eos_token (`str`, *optional*, defaults to `"<|endoftext|>"`):
The end of sequence token.
pad_token (`str`, *optional*, defaults to `"<|endoftext|>"`):
The token used for padding.
add_special_tokens (`bool`, *optional*, defaults to `True`):
Whether or not to add special tokens when encoding.
"""
model_input_names = ["input_ids", "attention_mask"]
def __init__(
self,
vocab_file,
merges_file,
errors="replace",
unk_token="<|endoftext|>",
bos_token=None,
eos_token="<|endoftext|>",
pad_token="<|endoftext|>",
add_prefix_space=False,
add_special_tokens=True,
**kwargs,
):
super().__init__(
vocab_file=vocab_file,
merges_file=merges_file,
errors=errors,
unk_token=unk_token,
bos_token=bos_token,
eos_token=eos_token,
pad_token=pad_token,
add_prefix_space=add_prefix_space,
add_special_tokens=add_special_tokens,
**kwargs,
)
# Add VibeVoice-specific special tokens
self._add_vibevoice_special_tokens()
def _add_vibevoice_special_tokens(self):
"""Add VibeVoice-specific special tokens."""
special_tokens = {
"additional_special_tokens": [
"<|vision_start|>", # Speech start (reusing vision tokens)
"<|vision_end|>", # Speech end
"<|vision_pad|>", # Speech diffusion pad
]
}
num_added = self.add_special_tokens(special_tokens)
# Cache special token IDs
self._speech_start_id = self.convert_tokens_to_ids("<|vision_start|>")
self._speech_end_id = self.convert_tokens_to_ids("<|vision_end|>")
self._speech_diffusion_id = self.convert_tokens_to_ids("<|vision_pad|>")
self._eos_id = self.convert_tokens_to_ids('<|endoftext|>')
return num_added
@property
def eos_id(self) -> int:
"""Id of the end of sequence token."""
return self._eos_id
@property
def speech_start_id(self) -> int:
"""Id of the speech start token."""
return self._speech_start_id
@property
def speech_end_id(self) -> int:
"""Id of the speech end token."""
return self._speech_end_id
@property
def speech_diffusion_id(self) -> int:
"""Id of the speech diffusion token."""
return self._speech_diffusion_id
@property
def pad_id(self) -> int:
"""Id used for padding (returns -100 for loss masking)."""
return -100
class VibeVoiceTextTokenizerFast(Qwen2TokenizerFast):
"""
Construct a "fast" VibeVoice tokenizer (backed by HuggingFace's *tokenizers* library).
Based on the Qwen2 tokenizer with additional special tokens for speech.
Args:
vocab_file (`str`, *optional*):
Path to the vocabulary file.
merges_file (`str`, *optional*):
Path to the merges file.
tokenizer_file (`str`, *optional*):
Path to [tokenizers](https://github.com/huggingface/tokenizers) file.
unk_token (`str`, *optional*, defaults to `"<|endoftext|>"`):
The unknown token.
bos_token (`str`, *optional*):
The beginning of sequence token. Not used for vibevoice.
eos_token (`str`, *optional*, defaults to `"<|endoftext|>"`):
The end of sequence token.
pad_token (`str`, *optional*, defaults to `"<|endoftext|>"`):
The token used for padding.
"""
model_input_names = ["input_ids", "attention_mask"]
def __init__(
self,
vocab_file=None,
merges_file=None,
tokenizer_file=None,
unk_token="<|endoftext|>",
bos_token=None,
eos_token="<|endoftext|>",
pad_token="<|endoftext|>",
add_prefix_space=False,
**kwargs,
):
super().__init__(
vocab_file=vocab_file,
merges_file=merges_file,
tokenizer_file=tokenizer_file,
unk_token=unk_token,
bos_token=bos_token,
eos_token=eos_token,
pad_token=pad_token,
add_prefix_space=add_prefix_space,
**kwargs,
)
# Add VibeVoice-specific special tokens
self._add_vibevoice_special_tokens()
def _add_vibevoice_special_tokens(self):
"""Add VibeVoice-specific special tokens."""
special_tokens = {
"additional_special_tokens": [
"<|vision_start|>", # Speech start (reusing vision tokens)
"<|vision_end|>", # Speech end
"<|vision_pad|>", # Speech diffusion pad
]
}
num_added = self.add_special_tokens(special_tokens)
# Cache special token IDs
self._speech_start_id = self.convert_tokens_to_ids("<|vision_start|>")
self._speech_end_id = self.convert_tokens_to_ids("<|vision_end|>")
self._speech_diffusion_id = self.convert_tokens_to_ids("<|vision_pad|>")
# self._eos_id = self.convert_tokens_to_ids('<|endoftext|>')
self._eos_id = self.eos_token_id # qwen2 / qwen3
self._pad_id = self.convert_tokens_to_ids('<|image_pad|>')
return num_added
@property
def eos_id(self) -> int:
"""Id of the end of sequence token."""
return self._eos_id
@property
def speech_start_id(self) -> int:
"""Id of the speech start token."""
return self._speech_start_id
@property
def speech_end_id(self) -> int:
"""Id of the speech end token."""
return self._speech_end_id
@property
def speech_diffusion_id(self) -> int:
"""Id of the speech diffusion token."""
return self._speech_diffusion_id
@property
def pad_id(self) -> int:
"""Id used for padding (returns -100 for loss masking)."""
return self._pad_id
__all__ = [
"VibeVoiceTextTokenizer",
"VibeVoiceTextTokenizerFast",
] |