AI Hardware & Software Stack

Which describes ONNX?

  1. AThe name of an AI-specific semiconductor chip
  2. BA method for compressing AI training data
  3. CA common model format for exchanging models between different frameworks and inference runtimes
Show answer and explanation

Answer: C. A common model format for exchanging models between different frameworks and inference runtimes

ONNX (Open Neural Network Exchange) is a standard for representing a model's computational graph and weights in a common format. A model trained in PyTorch or similar can be exported to ONNX and then loaded and run by other vendors' inference runtimes or hardware toolchains. Decoupling the training environment from the inference environment makes it easy to choose the best execution stack for each device.

  • A. It is not a chip name.
  • B. It is not a data compression method.
  • C. Correct. It stands for Open Neural Network Exchange.
Practice this topic Back to topic

More AI Hardware & Software Stack questions

  1. Why can inference runtimes such as TensorRT or ONNX Runtime run faster than using the training framework directly for inference?
  2. Why do element-wise operations such as ReLU or layer normalization tend to take a lot of time relative to their operation count in AI workloads?
  3. Why are convolutions in CNNs sometimes converted into matrix multiplications, for example with im2col?
  4. What is the difference between post-training quantization (PTQ) and quantization-aware training (QAT)?
  5. In post-training quantization of activations to INT8, what is the purpose of calibration?