Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

PyTorch Lightning cho bài kiểm tra đào tạo có tổ chức

Hiểu trách nhiệm của LightningModule và Trainer, lệnh gọi lại, giám sát điểm kiểm tra, chế độ tối ưu hóa và nhu cầu về các nguyên tắc cơ bản của PyTorch.

Đường dẫn hướng dẫn liên quanĐào tạo sét Pytorch
Câu hỏi 1 của 8

Which class typically defines model and task-specific steps in PyTorch Lightning?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. Which class typically defines model and task-specific steps in PyTorch Lightning?
    • Trainer
    • DataLoader worker
    • LightningModule
    • TensorBoard event file
  2. Which responsibility does the Trainer handle around task-specific model code?
    • Choosing labels from raw data
    • Orchestrating the training lifecycle and configured execution
    • Guaranteeing test-set accuracy
    • Defining every model layer automatically
  3. Why verify the metric name monitored by a checkpoint callback?
    • A missing monitor key can cause checkpoint selection to fail or monitor the wrong signal
    • It selects whether the Trainer uses CPU or GPU
    • It automatically normalizes the validation metric across devices
    • It chooses the model architecture
  4. When might manual optimization be appropriate?
    • When the algorithm needs custom control over optimizer steps
    • For every ordinary classifier
    • When no gradients are used
    • To remove validation
  5. Which claim about Lightning abstraction is accurate?
    • Lightning automatically decides the scientific evaluation metric
    • Lightning infers training labels from model outputs
    • PyTorch knowledge remains useful beneath training orchestration
    • Lightning removes the need for callback configuration
  6. What should be tested before scaling to distributed training?
    • A small run reveals device, metric and checkpoint behavior before costly scaling
    • A model title is enough to verify execution
    • Training-set loss alone validates distributed aggregation
    • A successful import proves the multi-device run is correct
  7. Why keep validation logic separate from the training objective?
    • To make callbacks optional
    • To reduce the number of model layers
    • To allow test labels into training
    • To evaluate without updating model weights on held-out data
  8. How can a team check that loop migration preserves training behavior?
    • Compare updates and metrics against a small reference run
    • Assume matching class names imply identical training behavior
    • Remove tests after conversion to Lightning
    • Compare checkpoint file sizes only