Manual Tuning

If the accuracy after sparsity does not meet the requirements, refer to this section to perform tuning.

If the accuracy of the retrained pruned model is not as expected, you can modify the simplified configuration file (see Simplified QAT Configuration File for details) and perform sparsity and training again. The common methods are as follows:

• Adjust the sparsity ratio, which is controlled by the prune_ratio parameter in the simplified configuration file. You can reduce the sparsity ratio and perform sparsity again for debugging.
• Skip certain layers in sparsity by setting the regular_prune_skip_layers parameter in the simplified configuration file.