6.6. FAQ and Common Failure Resolutions

This section includes the descriptions of Post-training Quantization (PTQ) FAQ and Common Failure Resolutions, which explain some common questions, troubleshooting phenomena and suggestions for dealing with them that you may encounter during the conversion process using PTQ.

• 6.6.1. Post-training Quantization (PTQ) FAQ
  • 6.6.1.1. How to understand the two forms of BPU acceleration and CPU computation mentioned in the operator constraint?
  • 6.6.1.2. How does model segmentation affect the performance?
  • 6.6.1.3. Why some OPs supported by BPU at the tail part of the model running on CPU?
  • 6.6.1.4. How to understand D-Robotics’s default calibration method?
  • 6.6.1.5. How to understand D-Robotics’s mix calibration method?
  • 6.6.1.6. How to understand the compiler optimization level parameters in yaml files?
  • 6.6.1.7. Why does the input size derived from the nv12 model hb_perf not match the prediction library?
  • 6.6.1.8. Is the data layout of the inputs to the quantized model and the on-board bin model necessarily the same?
  • 6.6.1.9. How to compile to get a multi-batch model?
  • 6.6.1.10. Is it normal for the order of model inputs to change during the conversion of a multi-input model?
• 6.6.2. Common Failure Resolutions
  • 6.6.2.1. Common Failure - hb_mapper checker
  • 6.6.2.2. Common Failure - hb_mapper makertbin
  • 6.6.2.3. Common Failure - hb_model_modifier
  • 6.6.2.4. Common Failure - hb_verifier
  • 6.6.2.5. Common Failure - hb_onnxruntime
  • 6.6.2.6. Common Failure - libDNN