Spaces:
Sleeping
Sleeping
File size: 2,779 Bytes
19ea5c5 | 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 | import os
import random
from dataclasses import dataclass
from typing import Optional
import numpy as np
import torch
@dataclass
class PathsConfig:
"""
Configuration for dataset and checkpoint paths.
This is tailored to your existing visually impaired dataset layout:
- Images: <data_root>/visual_dataset/*.jpg
- Text: <data_root>/visual_text/visual.token.txt
<data_root>/visual_text/visual.trainImages.txt
<data_root>/visual_text/visual.testImages.txt
"""
data_root: str = "/Users/ryan/Downloads/visuallyimpair"
images_dir_name: str = "visual_dataset"
text_dir_name: str = "visual_text"
def _join(self, *parts: str) -> str:
return os.path.join(*parts)
@property
def images_dir(self) -> str:
return self._join(self.data_root, self.images_dir_name)
@property
def text_dir(self) -> str:
return self._join(self.data_root, self.text_dir_name)
@property
def token_file(self) -> str:
return self._join(self.text_dir, "visual.token.txt")
@property
def train_list_file(self) -> str:
return self._join(self.text_dir, "visual.trainImages.txt")
@property
def test_list_file(self) -> str:
return self._join(self.text_dir, "visual.testImages.txt")
@dataclass
class TrainingConfig:
"""
Hyperparameters and training-related configuration.
"""
learning_rate: float = 5e-5
batch_size: int = 16
num_epochs: int = 10
warmup_steps: int = 500
max_caption_length: int = 50
gradient_accumulation_steps: int = 1
num_workers: int = 4
mixed_precision: bool = True
patience: int = 3
max_grad_norm: float = 1.0
# Model-specific
prefix_length: int = 1 # number of visual prefix tokens
# Logging / checkpoints
output_dir: str = "checkpoints"
log_dir: str = "runs"
# Reproducibility
seed: int = 42
def get_device() -> torch.device:
"""
Return the best available device (CUDA if available, else CPU) and log it.
"""
if torch.cuda.is_available():
device = torch.device("cuda")
print("Using CUDA for training/inference.")
else:
device = torch.device("cpu")
print("CUDA not available, falling back to CPU.")
return device
def set_seed(seed: int) -> None:
"""
Set random seeds for reproducibility across Python, NumPy, and PyTorch.
"""
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
def ensure_dir(path: str) -> None:
"""
Create directory if it does not already exist.
"""
os.makedirs(path, exist_ok=True)
|