Spaces:
Sleeping
Sleeping
File size: 9,093 Bytes
1ce5e18 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 |
# coding=utf-8
# Copyright 2023 HuggingFace Inc.
#
# 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.
import os
import pathlib
import tempfile
import uuid
import numpy as np
from ..utils import is_soundfile_availble, is_torch_available, is_vision_available, logging
logger = logging.get_logger(__name__)
if is_vision_available():
import PIL.Image
from PIL import Image
from PIL.Image import Image as ImageType
else:
ImageType = object
if is_torch_available():
import torch
if is_soundfile_availble():
import soundfile as sf
class AgentType:
"""
Abstract class to be reimplemented to define types that can be returned by agents.
These objects serve three purposes:
- They behave as they were the type they're meant to be, e.g., a string for text, a PIL.Image for images
- They can be stringified: str(object) in order to return a string defining the object
- They should be displayed correctly in ipython notebooks/colab/jupyter
"""
def __init__(self, value):
self._value = value
def __str__(self):
return self.to_string()
def to_raw(self):
logger.error(
"This is a raw AgentType of unknown type. Display in notebooks and string conversion will be unreliable"
)
return self._value
def to_string(self) -> str:
logger.error(
"This is a raw AgentType of unknown type. Display in notebooks and string conversion will be unreliable"
)
return str(self._value)
class AgentText(AgentType, str):
"""
Text type returned by the agent. Behaves as a string.
"""
def to_raw(self):
return self._value
def to_string(self):
return self._value
class AgentImage(AgentType, ImageType):
"""
Image type returned by the agent. Behaves as a PIL.Image.
"""
def __init__(self, value):
super().__init__(value)
if not is_vision_available():
raise ImportError("PIL must be installed in order to handle images.")
self._path = None
self._raw = None
self._tensor = None
if isinstance(value, ImageType):
self._raw = value
elif isinstance(value, (str, pathlib.Path)):
self._path = value
elif isinstance(value, torch.Tensor):
self._tensor = value
else:
raise ValueError(f"Unsupported type for {self.__class__.__name__}: {type(value)}")
def _ipython_display_(self, include=None, exclude=None):
"""
Displays correctly this type in an ipython notebook (ipython, colab, jupyter, ...)
"""
from IPython.display import Image, display
display(Image(self.to_string()))
def to_raw(self):
"""
Returns the "raw" version of that object. In the case of an AgentImage, it is a PIL.Image.
"""
if self._raw is not None:
return self._raw
if self._path is not None:
self._raw = Image.open(self._path)
return self._raw
def to_string(self):
"""
Returns the stringified version of that object. In the case of an AgentImage, it is a path to the serialized
version of the image.
"""
if self._path is not None:
return self._path
if self._raw is not None:
directory = tempfile.mkdtemp()
self._path = os.path.join(directory, str(uuid.uuid4()) + ".png")
self._raw.save(self._path)
return self._path
if self._tensor is not None:
array = self._tensor.cpu().detach().numpy()
# There is likely simpler than load into image into save
img = Image.fromarray((array * 255).astype(np.uint8))
directory = tempfile.mkdtemp()
self._path = os.path.join(directory, str(uuid.uuid4()) + ".png")
img.save(self._path)
return self._path
class AgentAudio(AgentType):
"""
Audio type returned by the agent.
"""
def __init__(self, value, samplerate=16_000):
super().__init__(value)
if not is_soundfile_availble():
raise ImportError("soundfile must be installed in order to handle audio.")
self._path = None
self._tensor = None
self.samplerate = samplerate
if isinstance(value, (str, pathlib.Path)):
self._path = value
elif isinstance(value, torch.Tensor):
self._tensor = value
else:
raise ValueError(f"Unsupported audio type: {type(value)}")
def _ipython_display_(self, include=None, exclude=None):
"""
Displays correctly this type in an ipython notebook (ipython, colab, jupyter, ...)
"""
from IPython.display import Audio, display
display(Audio(self.to_string(), rate=self.samplerate))
def to_raw(self):
"""
Returns the "raw" version of that object. It is a `torch.Tensor` object.
"""
if self._tensor is not None:
return self._tensor
if self._path is not None:
tensor, self.samplerate = sf.read(self._path)
self._tensor = torch.tensor(tensor)
return self._tensor
def to_string(self):
"""
Returns the stringified version of that object. In the case of an AgentAudio, it is a path to the serialized
version of the audio.
"""
if self._path is not None:
return self._path
if self._tensor is not None:
directory = tempfile.mkdtemp()
self._path = os.path.join(directory, str(uuid.uuid4()) + ".wav")
sf.write(self._path, self._tensor, samplerate=self.samplerate)
return self._path
AGENT_TYPE_MAPPING = {"text": AgentText, "image": AgentImage, "audio": AgentAudio}
INSTANCE_TYPE_MAPPING = {str: AgentText}
if is_vision_available():
INSTANCE_TYPE_MAPPING[PIL.Image] = AgentImage
def handle_agent_inputs(*args, **kwargs):
args = [(arg.to_raw() if isinstance(arg, AgentType) else arg) for arg in args]
kwargs = {k: (v.to_raw() if isinstance(v, AgentType) else v) for k, v in kwargs.items()}
return args, kwargs
def handle_agent_outputs(outputs, output_types=None):
if isinstance(outputs, dict):
decoded_outputs = {}
for i, (k, v) in enumerate(outputs.items()):
if output_types is not None:
# If the class has defined outputs, we can map directly according to the class definition
if output_types[i] in AGENT_TYPE_MAPPING:
decoded_outputs[k] = AGENT_TYPE_MAPPING[output_types[i]](v)
else:
decoded_outputs[k] = AgentType(v)
else:
# If the class does not have defined output, then we map according to the type
for _k, _v in INSTANCE_TYPE_MAPPING.items():
if isinstance(v, _k):
decoded_outputs[k] = _v(v)
if k not in decoded_outputs:
decoded_outputs[k] = AgentType[v]
elif isinstance(outputs, (list, tuple)):
decoded_outputs = type(outputs)()
for i, v in enumerate(outputs):
if output_types is not None:
# If the class has defined outputs, we can map directly according to the class definition
if output_types[i] in AGENT_TYPE_MAPPING:
decoded_outputs.append(AGENT_TYPE_MAPPING[output_types[i]](v))
else:
decoded_outputs.append(AgentType(v))
else:
# If the class does not have defined output, then we map according to the type
found = False
for _k, _v in INSTANCE_TYPE_MAPPING.items():
if isinstance(v, _k):
decoded_outputs.append(_v(v))
found = True
if not found:
decoded_outputs.append(AgentType(v))
else:
if output_types[0] in AGENT_TYPE_MAPPING:
# If the class has defined outputs, we can map directly according to the class definition
decoded_outputs = AGENT_TYPE_MAPPING[output_types[0]](outputs)
else:
# If the class does not have defined output, then we map according to the type
for _k, _v in INSTANCE_TYPE_MAPPING.items():
if isinstance(outputs, _k):
return _v(outputs)
return AgentType(outputs)
return decoded_outputs
|