File size: 7,395 Bytes
256a159 |
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 |
class VisualGLMMMBenchPromptConstructor:
"""MMBench prompt constructor for VisualGLM.
Args:
system_prompt (str): System prompt. (Default: '')
human_prompt (str): Human prompt. (Default: 'Q:')
assistant_prompt (str): Assistant prompt. (Default: 'A:')
"""
def __init__(self,
system_prompt: str = '',
human_prompt: str = 'Q:',
assistant_prompt: str = 'A:') -> None:
self.system_prompt = system_prompt
self.human_prompt = human_prompt
self.assistant_prompt = assistant_prompt
def __call__(self, batch: dict) -> tuple:
"""Construct prompt.
Args:
batch (dict): Input data containing image and data_samples.
Returns:
A tuple containing images, prompt, data_samples and image_position.
"""
assert len(batch['inputs']) == 1
image = batch.pop('inputs')[0].unsqueeze(0)
data_sample = batch.pop('data_samples')[0]
img_prompt = '<img></img>'
if data_sample.get('context') is not None:
prompt = img_prompt + self.system_prompt + self.human_prompt + data_sample.context + ' ' + data_sample.question + ' ' + data_sample.options # noqa
else:
prompt = img_prompt + self.system_prompt + self.human_prompt + data_sample.question + ' ' + data_sample.options # noqa
prompt += self.assistant_prompt
image_position = prompt.rfind('<img>') + 5
return image, prompt, data_sample, image_position
class VisualGLMBasePromptConstructor:
"""Base prompt constructor for VisualGLM.
The prompt will concat <img> and the given system prompt.
Args:
system_prompt (str): System prompt. (Default: '')
human_prompt (str): Human prompt. (Default: 'Q:')
assistant_prompt (str): Assistant prompt. (Default: 'A:')
"""
def __init__(self,
system_prompt: str = '',
human_prompt: str = 'Q:',
assistant_prompt: str = 'A:') -> None:
self.prompt = system_prompt
self.human_prompt = human_prompt
self.assistant_prompt = assistant_prompt
def __call__(self, batch: dict) -> tuple:
"""Construct prompt.
Args:
batch (dict): Input data containing image and data_samples.
Returns:
A tuple containing images, prompt, data_samples and image_position.
"""
assert len(batch['inputs']) == 1
image = batch.pop('inputs')[0].unsqueeze(0)
data_sample = batch.pop('data_samples')[0]
# generate text prompt
prompt = '<img></img>' + self.human_prompt + self.prompt + self.assistant_prompt # noqa
image_position = prompt.rfind('<img>') + 5
return image, prompt, data_sample, image_position
class VisualGLMVQAPromptConstructor(VisualGLMBasePromptConstructor):
"""VQA prompt constructor for VisualGLM.
The prompt will concat <img>, the question and the system prompt.
Args:
system_prompt (str): System prompt. (Default: '')
human_prompt (str): Human prompt. (Default: 'Q:')
assistant_prompt (str): Assistant prompt. (Default: 'A:')
"""
def __init__(self,
system_prompt='',
human_prompt: str = 'Q:',
assistant_prompt: str = 'A:') -> None:
super().__init__(system_prompt, human_prompt, assistant_prompt)
def __call__(self, batch: dict) -> tuple:
"""Construct prompt.
Args:
batch (dict): Input data containing image and data_samples.
Returns:
A tuple containing images, prompt, data_samples and image_position.
"""
assert len(batch['inputs']) == 1
image = batch.pop('inputs')[0].unsqueeze(0)
data_sample = batch.pop('data_samples')[0]
# generate text prompt
question = data_sample.get('question')
prompt = '<img></img>' + self.human_prompt + question + self.prompt
prompt += '\n' + self.assistant_prompt
image_position = prompt.rfind('<img>') + 5
return image, prompt, data_sample, image_position
class VisualGLMScienceQAPromptConstructor(VisualGLMBasePromptConstructor):
"""ScienceQA prompt constructor for VisualGLM.
The prompt will concat image and all terms in a question.
Args:
system_prompt (str): System prompt. (Default: '')
human_prompt (str): Human prompt. (Default: 'Q:')
assistant_prompt (str): Assistant prompt. (Default: 'A:')
"""
choice_mapping = {0: 'A', 1: 'B', 2: 'C', 3: 'D', 4: 'E', 5: 'F'}
def __init__(self,
system_prompt='',
human_prompt: str = 'Q:',
assistant_prompt: str = 'A:') -> None:
super().__init__(system_prompt, human_prompt, assistant_prompt)
def __call__(self, batch: dict) -> tuple:
"""Construct prompt.
Args:
batch (dict): Input data containing image and data_samples.
Returns:
A tuple containing images, prompt, data_samples and image_position.
"""
assert len(batch['inputs']) == 1
image = batch.pop('inputs')[0].unsqueeze(0)
data_sample = batch.pop('data_samples')[0]
questions = 'Question: ' + data_sample.get('question')
choices = data_sample.get('choices')
choices = [
f'({self.choice_mapping[i]}) ' + item
for i, item in enumerate(choices)
]
choices = 'Choices: ' + ' '.join(choices) + '\n'
contexts = 'Context: ' + data_sample.get('hint') + '\n'
# generate text prompt
prompt = '<img></img>' + self.human_prompt + contexts + questions + choices + self.prompt + self.assistant_prompt # noqa
image_position = prompt.rfind('<img>') + 5
return image, prompt, data_sample, image_position
class VisualGLMIconQAPromptConstructor(VisualGLMBasePromptConstructor):
"""IconQA prompt constructor for VisualGLM.
The prompt will concat <img>, the question and the system prompt.
Args:
system_prompt (str): System prompt. (Default: '')
human_prompt (str): Human prompt. (Default: 'Q:')
assistant_prompt (str): Assistant prompt. (Default: 'A:')
"""
def __init__(self,
system_prompt='',
human_prompt: str = 'Q:',
assistant_prompt: str = 'A:') -> None:
super().__init__(system_prompt, human_prompt, assistant_prompt)
def __call__(self, batch: dict) -> tuple:
"""Construct prompt.
Args:
batch (dict): Input data containing image and data_samples.
Returns:
A tuple containing images, prompt, data_samples and image_position.
"""
assert len(batch['inputs']) == 1
image = batch.pop('inputs')[0].unsqueeze(0)
data_sample = batch.pop('data_samples')[0]
questions = data_sample.get('question') + '\n'
choices = data_sample.get('choices')
choices = 'Options: ' + ', '.join(choices) + '.\n'
# generate text prompt
prompt = '<img></img>' + self.human_prompt + questions + choices + self.prompt + self.assistant_prompt # noqa
image_position = prompt.rfind('<img>') + 5
return image, prompt, data_sample, image_position
|