================================================================================
Model Configuration (CRITICAL FOR INFERENCE)
================================================================================
modelCategory: KD_NAS
Task: classification
Genome: [[[5, 1], [0, 1]], [[4, 1], [2, 1]], [[3, 2], [3, 0]], [[5, 3], [3, 0]], [[4, 5], [1, 0]]]
Architecture: KDAdapterGenotype(ops=[('dw_conv_3x3', 1), ('skip_connect', 1), ('conv_3x3', 1), ('avg_pool_3x3', 1), ('conv_1x1', 2), ('conv_1x1', 0), ('dw_conv_3x3', 3), ('conv_1x1', 0), ('conv_3x3', 5), ('max_pool_3x3', 0)], ops_concat=[4, 6])
kd_teacher: yolov8n
search_space: micro
num_cells: 1
num_classes: 10
image_size: 224
================================================================================
Performance Metrics
================================================================================
Params: 0.10708 MB
Forward FLOPs: 1351.140032 M
Training FLOPs: 1351.140032 M [approx]
--------------------------------------------------------------------------------
Epoch  | Train Acc  | Valid Acc  | Error      | FLOPs     
--------------------------------------------------------------------------------
1      | 60.69      | 61.38      | 38.62      | 1351.14   
2      | 63.57      | 64.74      | 35.26      | 1351.14   

==================================================
Pareto Front (Search Results)
==================================================
selection_method = knee_point
total_architectures_evaluated = 9

Top architectures:
  1. acc=60.9900, params=0.1071
  2. acc=55.9100, params=0.0343
  3. acc=49.7600, params=0.0337
