Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ
TorchServe dành cho bài kiểm tra mô hình PyTorch
Hiểu các kho lưu trữ, trình xử lý và điều chỉnh nhân viên của TorchServe trong khi tính toán trạng thái bảo trì hiện tại của dự án.
Câu hỏi 1 của 8
What does a TorchServe handler commonly define?
Xem trước mọi câu hỏi8 câu hỏi
Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.
- What does a TorchServe handler commonly define?
- Request preprocessing, model inference and response formatting
- The cloud provider's GPU quota
- The model's training labels only
- A dashboard alert threshold
- What can a model archive package?
- The physical GPU kernel driver
- Weights and files or metadata needed to serve a model
- Every user's request history
- Only a Docker image tag
- How can increasing worker count affect GPU serving?
- It removes the need for batching
- It always lowers latency without tradeoffs
- It can increase concurrency and also increase memory use
- It changes model weights
- Which compromise can dynamic batching introduce?
- It eliminates queueing
- It guarantees the same response time for all requests
- It reduces batch size to one
- Higher throughput may come with added waiting latency
- What does TorchServe's current upstream notice say?
- Every deployment is automatically supported
- The project is in limited maintenance and no longer actively maintained
- The project requires no security updates
- The model archive format is a GPU driver
- Why test a custom handler independently?
- It determines the model's training data
- It changes the CUDA driver automatically
- It has its own preprocessing and response-contract behavior
- It replaces vulnerability review
- Why review access to management endpoints?
- They contain only static documentation
- They are required for every public client
- They provide a guarantee of accuracy
- They can expose operational controls and should be restricted
- What should a new deployment decision consider beyond feature fit?
- Long-term maintenance, security fixes and migration ownership
- Only whether a demo runs
- The number of handler lines
- The model's training accuracy alone