Spaces:
Sleeping
Sleeping
File size: 9,465 Bytes
8d272fe |
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 |
"""
Configuration for data processing in LLaVA.
"""
from dataclasses import dataclass, field
from typing import Optional, List, Dict, Any
@dataclass
class DatasetConfig:
"""Configuration for a dataset."""
name: str
path: str
image_folder: str
annotation_file: str
split: str = "train"
max_samples: int = -1 # -1 means use all samples
image_key: str = "image"
caption_key: str = "caption"
question_key: str = "question"
answer_key: str = "answer"
@dataclass
class PretrainingDataConfig:
"""Configuration for pretraining data."""
datasets: List[DatasetConfig] = field(default_factory=list)
image_size: int = 336
max_length: int = 2048
pad_to_max_length: bool = False
preprocessing_num_workers: int = 4
overwrite_cache: bool = False
@classmethod
def default(cls) -> "PretrainingDataConfig":
"""Create a default pretraining data configuration."""
return cls(
datasets=[
DatasetConfig(
name="laion-cc-sbu-558k",
path="liuhaotian/LLaVA-Pretrain",
image_folder="",
annotation_file="blip_laion_cc_sbu_558k.json",
image_key="image",
caption_key="caption"
)
]
)
@dataclass
class InstructionTuningDataConfig:
"""Configuration for instruction tuning data."""
datasets: List[DatasetConfig] = field(default_factory=list)
image_size: int = 336
max_length: int = 2048
pad_to_max_length: bool = False
preprocessing_num_workers: int = 4
overwrite_cache: bool = False
@classmethod
def default(cls) -> "InstructionTuningDataConfig":
"""Create a default instruction tuning data configuration."""
return cls(
datasets=[
DatasetConfig(
name="llava-instruct-150k",
path="liuhaotian/LLaVA-Instruct-150K",
image_folder="",
annotation_file="llava_instruct_150k.json",
image_key="image",
question_key="question",
answer_key="answer"
)
]
)
@dataclass
class EvaluationDataConfig:
"""Configuration for evaluation data."""
datasets: List[DatasetConfig] = field(default_factory=list)
image_size: int = 336
max_length: int = 2048
pad_to_max_length: bool = False
preprocessing_num_workers: int = 4
@classmethod
def default(cls) -> "EvaluationDataConfig":
"""Create a default evaluation data configuration."""
return cls(
datasets=[
DatasetConfig(
name="vqav2",
path="",
image_folder="coco/val2014",
annotation_file="vqav2_val.json",
split="val",
question_key="question",
answer_key="answer"
),
DatasetConfig(
name="gqa",
path="",
image_folder="gqa/images",
annotation_file="gqa_testdev_balanced.json",
split="testdev",
question_key="question",
answer_key="answer"
),
DatasetConfig(
name="textvqa",
path="",
image_folder="textvqa/train_images",
annotation_file="textvqa_val.json",
split="val",
question_key="question",
answer_key="answer"
)
]
)
@dataclass
class DataConfig:
"""Configuration for all data processing."""
pretraining: PretrainingDataConfig = field(default_factory=PretrainingDataConfig.default)
instruction_tuning: InstructionTuningDataConfig = field(default_factory=InstructionTuningDataConfig.default)
evaluation: EvaluationDataConfig = field(default_factory=EvaluationDataConfig.default)
@classmethod
def from_dict(cls, config_dict: Dict[str, Any]) -> "DataConfig":
"""Create a configuration from a dictionary."""
# Process pretraining config
pretraining_dict = config_dict.get("pretraining", {})
pretraining_datasets = []
for dataset_dict in pretraining_dict.get("datasets", []):
pretraining_datasets.append(DatasetConfig(**dataset_dict))
pretraining_config = PretrainingDataConfig(
datasets=pretraining_datasets,
**{k: v for k, v in pretraining_dict.items() if k != "datasets"}
)
# Process instruction tuning config
instruction_dict = config_dict.get("instruction_tuning", {})
instruction_datasets = []
for dataset_dict in instruction_dict.get("datasets", []):
instruction_datasets.append(DatasetConfig(**dataset_dict))
instruction_config = InstructionTuningDataConfig(
datasets=instruction_datasets,
**{k: v for k, v in instruction_dict.items() if k != "datasets"}
)
# Process evaluation config
evaluation_dict = config_dict.get("evaluation", {})
evaluation_datasets = []
for dataset_dict in evaluation_dict.get("datasets", []):
evaluation_datasets.append(DatasetConfig(**dataset_dict))
evaluation_config = EvaluationDataConfig(
datasets=evaluation_datasets,
**{k: v for k, v in evaluation_dict.items() if k != "datasets"}
)
# Create and return the configuration
return cls(
pretraining=pretraining_config,
instruction_tuning=instruction_config,
evaluation=evaluation_config
)
def to_dict(self) -> Dict[str, Any]:
"""Convert the configuration to a dictionary."""
# Process pretraining datasets
pretraining_datasets = []
for dataset in self.pretraining.datasets:
pretraining_datasets.append({
"name": dataset.name,
"path": dataset.path,
"image_folder": dataset.image_folder,
"annotation_file": dataset.annotation_file,
"split": dataset.split,
"max_samples": dataset.max_samples,
"image_key": dataset.image_key,
"caption_key": dataset.caption_key,
"question_key": dataset.question_key,
"answer_key": dataset.answer_key
})
# Process instruction tuning datasets
instruction_datasets = []
for dataset in self.instruction_tuning.datasets:
instruction_datasets.append({
"name": dataset.name,
"path": dataset.path,
"image_folder": dataset.image_folder,
"annotation_file": dataset.annotation_file,
"split": dataset.split,
"max_samples": dataset.max_samples,
"image_key": dataset.image_key,
"caption_key": dataset.caption_key,
"question_key": dataset.question_key,
"answer_key": dataset.answer_key
})
# Process evaluation datasets
evaluation_datasets = []
for dataset in self.evaluation.datasets:
evaluation_datasets.append({
"name": dataset.name,
"path": dataset.path,
"image_folder": dataset.image_folder,
"annotation_file": dataset.annotation_file,
"split": dataset.split,
"max_samples": dataset.max_samples,
"image_key": dataset.image_key,
"caption_key": dataset.caption_key,
"question_key": dataset.question_key,
"answer_key": dataset.answer_key
})
# Create the configuration dictionary
config_dict = {
"pretraining": {
"datasets": pretraining_datasets,
"image_size": self.pretraining.image_size,
"max_length": self.pretraining.max_length,
"pad_to_max_length": self.pretraining.pad_to_max_length,
"preprocessing_num_workers": self.pretraining.preprocessing_num_workers,
"overwrite_cache": self.pretraining.overwrite_cache
},
"instruction_tuning": {
"datasets": instruction_datasets,
"image_size": self.instruction_tuning.image_size,
"max_length": self.instruction_tuning.max_length,
"pad_to_max_length": self.instruction_tuning.pad_to_max_length,
"preprocessing_num_workers": self.instruction_tuning.preprocessing_num_workers,
"overwrite_cache": self.instruction_tuning.overwrite_cache
},
"evaluation": {
"datasets": evaluation_datasets,
"image_size": self.evaluation.image_size,
"max_length": self.evaluation.max_length,
"pad_to_max_length": self.evaluation.pad_to_max_length,
"preprocessing_num_workers": self.evaluation.preprocessing_num_workers
}
}
return config_dict |