File size: 12,192 Bytes
d01bd6b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 |
# coding=utf-8
# Copyright 2024 The Qwen team, Alibaba Group and the HuggingFace Inc. team. All rights reserved.
#
# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
# and OPT implementations in this library. It has been modified from its
# original forms to accommodate minor architectural differences compared
# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Processor class for EMOVA with qwen2vit.
"""
import json
from typing import List, Union
from transformers import AutoProcessor, AutoImageProcessor
try:
from typing import Unpack
except ImportError:
from typing_extensions import Unpack
from transformers.feature_extraction_utils import BatchFeature
from .image_utils import ImageInput, VideoInput
from transformers.processing_utils import (
ProcessingKwargs,
ProcessorMixin,
)
from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
from transformers.utils import logging
from .configuration_emova import EMOVAConfig
from .image_processing_emova import EMOVAImageProcessor
logger = logging.get_logger(__name__)
class EMOVAProcessorKwargs(ProcessingKwargs, total=False):
_defaults = {
"text_kwargs": {
"padding": False,
},
}
class EMOVAProcessor(ProcessorMixin):
r"""
Constructs a Qwen2-VL processor which wraps a Qwen2-VL image processor and a Qwen2 tokenizer into a single processor.
[`EMOVAProcessor`] offers all the functionalities of [`EmovaImageProcessor`] and [`Qwen2TokenizerFast`]. See the
[`~EMOVAProcessor.__call__`] and [`~EMOVAProcessor.decode`] for more information.
Args:
image_processor ([`EmovaImageProcessor`], *optional*):
The image processor is a required input.
tokenizer ([`Qwen2TokenizerFast`], *optional*):
The tokenizer is a required input.
chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages
in a chat into a tokenizable string.
"""
attributes = ["image_processor", "tokenizer"]
valid_kwargs = ["chat_template"]
image_processor_class = "AutoImageProcessor"
# image_processor_class = "EMOVAImageProcessor"
tokenizer_class = ("Qwen2Tokenizer", "Qwen2TokenizerFast")
def __init__(self, image_processor=None, tokenizer=None, chat_template=None, **kwargs):
super().__init__(image_processor=image_processor, tokenizer=tokenizer, chat_template=chat_template)
self.speech_tokenizer = None
def set_speech_tokenizer(self, tokenizer=None):
if self.speech_tokenizer and tokenizer:
logger.info('You are resetting speech tokenizer!')
return
self.speech_tokenizer = tokenizer
logger.info('Setting speech tokenizer!')
def prepare_audio_input(self, text, audio, has_image=False):
if text[0]["role"] == "system":
system_prompt = text[0]
valid_index = 1
else:
system_prompt = None
valid_index = 0
logger.warning("Audio inputs are given, but system prompts are not given.")
if len(text) > valid_index:
logger.warning("When audio inputs are given, text inputs except system prompts will be discarded.")
audio_chat_format = r'Please recognize the texts, emotion and pitch from the user question speech units and provide the texts, emotion, pitch and speech units for the assistant response. \nEmotion should be chosen from ["neutral", "happy", "sad", "angry", "surprised", "disgusted", "fearful"]. \nPitch should be chosen from ["low", "normal", "high"].\nYour output should be in json format.\nAn output example is:\n{"user question text": "", "user question emotion": "", "user question pitch": "", "assistant response text": "", "assistant response emotion": "", "assistant response pitch": ""，"assistant response speech": ""}\n\nuser question speech:'
audio_chat_prompt = audio_chat_format + self.speech_tokenizer.encode(audio)
if has_image:
audio_chat_input = {
"role": "user",
"content": [{"type": "image"}, {"type": "text", "text": audio_chat_prompt}],
}
else:
audio_chat_input = {
"role": "user",
"content": [{"type": "text", "text": audio_chat_prompt}],
}
return [system_prompt, audio_chat_input] if system_prompt else [audio_chat_input]
def prepare_audio_output(self, output):
try:
if output.startswith('{"{"'):
return self.prepare_audio_output(output[2:])
if output.startswith("{"):
if output.endswith("|>"):
output += "\"}"
elif output.endswith("\""):
output += "}"
info_dict = json.loads(output)
content_unit = info_dict['assistant response speech'].strip()
emotion = info_dict['assistant response emotion'] if 'assistant response emotion' in info_dict else "neutral"
speed = info_dict['assistant response speed'] if 'assistant response speed' in info_dict else "normal"
pitch = info_dict['assistant response pitch'] if 'assistant response pitch' in info_dict else "normal"
except:
content_unit = output.strip()
emotion = 'neutral'
speed = "normal"
pitch = "normal"
return content_unit, emotion, speed, pitch
def __call__(
self,
text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
images: ImageInput = None,
audios: Union[str, List[str]] = None,
**kwargs: Unpack[EMOVAProcessorKwargs],
) -> BatchFeature:
"""
Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
and `kwargs` arguments to Qwen2TokenizerFast's [`~Qwen2TokenizerFast.__call__`] if `text` is not `None` to encode
the text. To prepare the vision inputs, this method forwards the `vision_infos` and `kwrags` arguments to
EmovaImageProcessor's [`~EmovaImageProcessor.__call__`] if `vision_infos` is not `None`.
Args:
images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
tensor. Both channels-first and channels-last formats are supported.
text (`str`, `List[str]`, `List[List[str]]`):
The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
(pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
`is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
audios (`str`, `List[str]`): Paths to the audio input(s).
videos (`np.ndarray`, `torch.Tensor`, `List[np.ndarray]`, `List[torch.Tensor]`):
The image or batch of videos to be prepared. Each video can be a 4D NumPy array or PyTorch
tensor, or a nested list of 3D frames. Both channels-first and channels-last formats are supported.
return_tensors (`str` or [`~utils.TensorType`], *optional*):
If set, will return tensors of a particular framework. Acceptable values are:
- `'tf'`: Return TensorFlow `tf.constant` objects.
- `'pt'`: Return PyTorch `torch.Tensor` objects.
- `'np'`: Return NumPy `np.ndarray` objects.
- `'jax'`: Return JAX `jnp.ndarray` objects.
Returns:
[`BatchFeature`]: A [`BatchFeature`] with the following fields:
- **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
- **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
`return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
`None`).
- **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
- **pixel_values_videos** -- Pixel values of videos to be fed to a model. Returned when `videos` is not `None`.
- **image_grid_thw** -- List of image 3D grid in LLM. Returned when `images` is not `None`.
- **video_grid_thw** -- List of video 3D grid in LLM. Returned when `videos` is not `None`.
"""
output_kwargs = self._merge_kwargs(
EMOVAProcessorKwargs,
tokenizer_init_kwargs=self.tokenizer.init_kwargs,
**kwargs,
)
if images is not None:
image_inputs = self.image_processor(images=images, videos=None, **output_kwargs["images_kwargs"])
image_grid_thw = image_inputs.pop("image_grid_thw")
image_inputs['image_sizes'] = image_grid_thw
else:
image_inputs = {}
image_sizes = None
if audios is not None:
audios = [audios] if not isinstance(audios, list) else audios
text = [text] if not isinstance(text[0], list) else text
assert len(audios) == len(text), "Audio inputs should correspond with text inputs."
assert self.speech_tokenizer, "Audio inputs are given, while speech tokenizer is not set. Call `EMOVAProcessor.prepare_audio_input()` before processing audio inputs."
text = [self.prepare_audio_input(txt, audio, has_image=images is not None) for txt, audio in zip(text, audios)]
if not isinstance(text, list):
text = [text]
_ = output_kwargs["text_kwargs"].pop("padding_side", None)
try:
text = self.apply_chat_template(text, add_generation_prompt=True, padding=True)
except Exception as e:
logger.info('Warning: input texts have been applied chat templates!')
text_inputs = self.tokenizer(text, **output_kwargs["text_kwargs"])
return BatchFeature(data={**text_inputs, **image_inputs})
def batch_decode(self, sequences, output_wav_prefix='output', *args, **kwargs):
return [self.decode(seq, output_wav_file="{}_{}.wav".format(output_wav_prefix, i), *args, **kwargs)
for i, seq in enumerate(sequences)]
def decode(self, *args, speaker='female', output_wav_file='output.wav', **kwargs):
output = self.tokenizer.decode(*args, **kwargs)
if '<|speech_' not in output:
return output
content_unit, emotion, speed, pitch = self.prepare_audio_output(output)
gender = speaker.lower()
condition = f'gender-{gender}_emotion-{emotion}_speed-{speed}_pitch-{pitch}'
self.speech_tokenizer.decode(content_unit, condition=condition, output_wav_file=output_wav_file)
return output_wav_file
@property
def model_input_names(self):
tokenizer_input_names = self.tokenizer.model_input_names
image_processor_input_names = self.image_processor.model_input_names
return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
|