yolov3_mobilenetv1_prune_qat.yml 1014 B

12345678910111213141516171819
  1. # Weights of yolov3_mobilenet_v1_voc
  2. pretrain_weights: https://paddledet.bj.bcebos.com/models/yolov3_mobilenet_v1_270e_voc.pdparams
  3. slim: PrunerQAT
  4. PrunerQAT:
  5. criterion: fpgm
  6. pruned_params: ['conv2d_27.w_0', 'conv2d_28.w_0', 'conv2d_29.w_0',
  7. 'conv2d_30.w_0', 'conv2d_31.w_0', 'conv2d_32.w_0',
  8. 'conv2d_34.w_0', 'conv2d_35.w_0', 'conv2d_36.w_0',
  9. 'conv2d_37.w_0', 'conv2d_38.w_0', 'conv2d_39.w_0',
  10. 'conv2d_41.w_0', 'conv2d_42.w_0', 'conv2d_43.w_0',
  11. 'conv2d_44.w_0', 'conv2d_45.w_0', 'conv2d_46.w_0']
  12. pruned_ratios: [0.1,0.2,0.2,0.2,0.2,0.1,0.2,0.3,0.3,0.3,0.2,0.1,0.3,0.4,0.4,0.4,0.4,0.3]
  13. print_prune_params: False
  14. quant_config: {
  15. 'weight_quantize_type': 'channel_wise_abs_max', 'activation_quantize_type': 'moving_average_abs_max',
  16. 'weight_bits': 8, 'activation_bits': 8, 'dtype': 'int8', 'window_size': 10000, 'moving_rate': 0.9,
  17. 'quantizable_layer_type': ['Conv2D', 'Linear']}
  18. print_qat_model: True