AMP: true
AUG:
  AUTO_AUGMENT: rand-m9-mstd0.5-inc1
  COLOR_JITTER: 0.4
  CUTMIX: 1.0
  CUTMIX_MINMAX: null
  MESA: 1.0
  MIXUP: 0.8
  MIXUP_MODE: batch
  MIXUP_PROB: 1.0
  MIXUP_SWITCH_PROB: 0.5
  RECOUNT: 1
  REMODE: pixel
  REPROB: 0.25
BASE:
- ''
DATA:
  BATCH_SIZE: 256
  CACHE_MODE: part
  DATASET: imagenet
  DATA_PATH: /data/imagenet
  IMG_SIZE: 256
  INTERPOLATION: bicubic
  NUM_WORKERS: 16
  PIN_MEMORY: true
  ZIP_MODE: false
EVAL_MODE: false
LOCAL_RANK: 0
MODEL:
  DROP_PATH_RATE: 0.2
  DROP_RATE: 0.0
  LABEL_SMOOTHING: 0.1
  MLLA:
    APE: false
    DEPTHS:
    - 2
    - 4
    - 8
    - 4
    EMBED_DIM: 48
    EXPANSION_RATIO: 2.0
    IN_CHANS: 3
    MLP_RATIO: 4.0
    NUM_HEADS:
    - 2
    - 4
    - 8
    - 16
    PATCH_SIZE: 4
    QKV_BIAS: true
  NAME: mlla_mini
  NUM_CLASSES: 1000
  RESUME: ''
  TYPE: mlla
OUTPUT: output/300e/mlla_mini/recattn_nearest_interp_simplify
PRINT_FREQ: 100
SAVE_FREQ: 1
SEED: 0
TAG: recattn_nearest_interp_simplify
TEST:
  CROP: true
THROUGHPUT_MODE: false
TRAIN:
  AUTO_RESUME: true
  BASE_LR: 0.001
  CLIP_GRAD: 5.0
  COOLDOWN_EPOCHS: 0
  EPOCHS: 300
  LR_SCHEDULER:
    DECAY_EPOCHS: 30
    DECAY_RATE: 0.1
    NAME: cosine
  MIN_LR: 1.0e-05
  OPTIMIZER:
    BETAS:
    - 0.9
    - 0.999
    EPS: 1.0e-08
    MOMENTUM: 0.9
    NAME: adamw
  START_EPOCH: 0
  USE_CHECKPOINT: false
  WARMUP_EPOCHS: 20
  WARMUP_LR: 1.0e-06
  WEIGHT_DECAY: 0.05

Creating model:mlla/mlla_mini
MLLA(
  (patch_embed): Stem(
    (conv1): ConvLayer(
      (conv): Conv2d(3, 24, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (norm): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (act): ReLU()
    )
    (conv2): Sequential(
      (0): ConvLayer(
        (conv): Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (norm): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (act): ReLU()
      )
      (1): ConvLayer(
        (conv): Conv2d(24, 24, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
        (norm): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (conv3): Sequential(
      (0): ConvLayer(
        (conv): Conv2d(24, 192, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
        (norm): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (act): ReLU()
      )
      (1): ConvLayer(
        (conv): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (norm): BatchNorm2d(48, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
  )
  (pos_drop): Dropout(p=0.0, inplace=False)
  (layers): ModuleList(
    (0): BasicLayer(
      dim=48, level=1, depth=2
      (blocks): ModuleList(
        (0): MLLABlock(
          dim=48, mlp_ratio=4.0, level=1, downsample=False
          (cpe1): Conv2d(48, 48, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=48)
          (norm1): LayerNorm((48,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=48, out_features=96, bias=True)
          (o_proj): Conv2d(48, 48, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(48, 48, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=48)
              (1): LinearAttention(
                dim=48, num_heads=2
                (qk): Conv2d(48, 96, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=48)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(48, 48, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=48)
          )
          (cpe2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=48)
          (norm2): LayerNorm((48,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=48, out_features=192, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=192, out_features=48, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): Identity()
        )
        (1): MLLABlock(
          dim=48, mlp_ratio=4.0, level=1, downsample=False
          (cpe1): Conv2d(48, 48, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=48)
          (norm1): LayerNorm((48,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=48, out_features=96, bias=True)
          (o_proj): Conv2d(48, 48, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(48, 48, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=48)
              (1): LinearAttention(
                dim=48, num_heads=2
                (qk): Conv2d(48, 96, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=48)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(48, 48, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=48)
          )
          (cpe2): Conv2d(48, 48, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=48)
          (norm2): LayerNorm((48,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=48, out_features=192, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=192, out_features=48, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.012)
        )
      )
      (downsample): MLLABlock(
        dim=48, mlp_ratio=4.0, level=0, downsample=True
        (cpe1): Conv2d(48, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=48)
        (norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
        (i_proj): Linear(in_features=96, out_features=192, bias=True)
        (o_proj): Conv2d(96, 96, kernel_size=(1, 1), stride=(1, 1))
        (act): SiLU()
        (agg): RecAttn2d(
          (down): Sequential(
            (0): Conv2d(96, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96)
            (1): LinearAttention(
              dim=96, num_heads=2
              (qk): Conv2d(96, 192, kernel_size=(1, 1), stride=(1, 1), groups=2)
              (elu): ELU(alpha=1.0)
              (lepe): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
            )
            (2): Upsample(scale_factor=2.0, mode='nearest')
          )
          (conv): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
        )
        (cpe2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
        (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=96, out_features=384, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=384, out_features=96, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): DropPath(drop_prob=0.012)
      )
    )
    (1): BasicLayer(
      dim=96, level=0, depth=4
      (blocks): ModuleList(
        (0): MLLABlock(
          dim=96, mlp_ratio=4.0, level=0, downsample=False
          (cpe1): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          (norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=96, out_features=192, bias=True)
          (o_proj): Conv2d(96, 96, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(96, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96)
              (1): LinearAttention(
                dim=96, num_heads=4
                (qk): Conv2d(96, 192, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          )
          (cpe2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
          (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=96, out_features=384, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=384, out_features=96, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.024)
        )
        (1): MLLABlock(
          dim=96, mlp_ratio=4.0, level=0, downsample=False
          (cpe1): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          (norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=96, out_features=192, bias=True)
          (o_proj): Conv2d(96, 96, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(96, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96)
              (1): LinearAttention(
                dim=96, num_heads=4
                (qk): Conv2d(96, 192, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          )
          (cpe2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
          (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=96, out_features=384, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=384, out_features=96, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.035)
        )
        (2): MLLABlock(
          dim=96, mlp_ratio=4.0, level=0, downsample=False
          (cpe1): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          (norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=96, out_features=192, bias=True)
          (o_proj): Conv2d(96, 96, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(96, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96)
              (1): LinearAttention(
                dim=96, num_heads=4
                (qk): Conv2d(96, 192, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          )
          (cpe2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
          (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=96, out_features=384, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=384, out_features=96, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.047)
        )
        (3): MLLABlock(
          dim=96, mlp_ratio=4.0, level=0, downsample=False
          (cpe1): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          (norm1): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=96, out_features=192, bias=True)
          (o_proj): Conv2d(96, 96, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(96, 96, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96)
              (1): LinearAttention(
                dim=96, num_heads=4
                (qk): Conv2d(96, 192, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(96, 96, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=96)
          )
          (cpe2): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=96)
          (norm2): LayerNorm((96,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=96, out_features=384, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=384, out_features=96, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.059)
        )
      )
      (downsample): MLLABlock(
        dim=96, mlp_ratio=4.0, level=-1, downsample=True
        (cpe1): Conv2d(96, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=96)
        (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
        (i_proj): Linear(in_features=192, out_features=384, bias=True)
        (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
        (act): SiLU()
        (agg): RecAttn2d(
          (down): Sequential(
            (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
            (1): LinearAttention(
              dim=192, num_heads=4
              (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
              (elu): ELU(alpha=1.0)
              (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
            )
            (2): Upsample(scale_factor=2.0, mode='nearest')
          )
          (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
        )
        (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
        (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=192, out_features=768, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=768, out_features=192, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): DropPath(drop_prob=0.059)
      )
    )
    (2): BasicLayer(
      dim=192, level=-1, depth=8
      (blocks): ModuleList(
        (0): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.071)
        )
        (1): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.082)
        )
        (2): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.094)
        )
        (3): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.106)
        )
        (4): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.118)
        )
        (5): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.129)
        )
        (6): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.141)
        )
        (7): MLLABlock(
          dim=192, mlp_ratio=4.0, level=-1, downsample=False
          (cpe1): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          (norm1): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=192, out_features=384, bias=True)
          (o_proj): Conv2d(192, 192, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(192, 192, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
              (1): LinearAttention(
                dim=192, num_heads=8
                (qk): Conv2d(192, 384, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(192, 192, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=192)
          )
          (cpe2): Conv2d(192, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=192)
          (norm2): LayerNorm((192,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=192, out_features=768, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=768, out_features=192, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.153)
        )
      )
      (downsample): MLLABlock(
        dim=192, mlp_ratio=4.0, level=-2, downsample=True
        (cpe1): Conv2d(192, 384, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=192)
        (norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
        (i_proj): Linear(in_features=384, out_features=768, bias=True)
        (o_proj): Conv2d(384, 384, kernel_size=(1, 1), stride=(1, 1))
        (act): SiLU()
        (agg): RecAttn2d(
          (down): Sequential(
            (0): Conv2d(384, 384, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=384)
            (1): LinearAttention(
              dim=384, num_heads=8
              (qk): Conv2d(384, 768, kernel_size=(1, 1), stride=(1, 1), groups=2)
              (elu): ELU(alpha=1.0)
              (lepe): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
            )
            (2): Upsample(scale_factor=2.0, mode='nearest')
          )
          (conv): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
        )
        (cpe2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
        (norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
        (mlp): Mlp(
          (fc1): Linear(in_features=384, out_features=1536, bias=True)
          (act): GELU(approximate='none')
          (fc2): Linear(in_features=1536, out_features=384, bias=True)
          (drop): Dropout(p=0.0, inplace=False)
        )
        (drop_path): DropPath(drop_prob=0.153)
      )
    )
    (3): BasicLayer(
      dim=384, level=-2, depth=4
      (blocks): ModuleList(
        (0): MLLABlock(
          dim=384, mlp_ratio=4.0, level=-2, downsample=False
          (cpe1): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          (norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=384, out_features=768, bias=True)
          (o_proj): Conv2d(384, 384, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(384, 384, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=384)
              (1): LinearAttention(
                dim=384, num_heads=16
                (qk): Conv2d(384, 768, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          )
          (cpe2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
          (norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=384, out_features=1536, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=1536, out_features=384, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.165)
        )
        (1): MLLABlock(
          dim=384, mlp_ratio=4.0, level=-2, downsample=False
          (cpe1): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          (norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=384, out_features=768, bias=True)
          (o_proj): Conv2d(384, 384, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(384, 384, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=384)
              (1): LinearAttention(
                dim=384, num_heads=16
                (qk): Conv2d(384, 768, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          )
          (cpe2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
          (norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=384, out_features=1536, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=1536, out_features=384, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.176)
        )
        (2): MLLABlock(
          dim=384, mlp_ratio=4.0, level=-2, downsample=False
          (cpe1): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          (norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=384, out_features=768, bias=True)
          (o_proj): Conv2d(384, 384, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(384, 384, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=384)
              (1): LinearAttention(
                dim=384, num_heads=16
                (qk): Conv2d(384, 768, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          )
          (cpe2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
          (norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=384, out_features=1536, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=1536, out_features=384, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.188)
        )
        (3): MLLABlock(
          dim=384, mlp_ratio=4.0, level=-2, downsample=False
          (cpe1): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          (norm1): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (i_proj): Linear(in_features=384, out_features=768, bias=True)
          (o_proj): Conv2d(384, 384, kernel_size=(1, 1), stride=(1, 1))
          (act): SiLU()
          (agg): RecAttn2d(
            (down): Sequential(
              (0): Conv2d(384, 384, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=384)
              (1): LinearAttention(
                dim=384, num_heads=16
                (qk): Conv2d(384, 768, kernel_size=(1, 1), stride=(1, 1), groups=2)
                (elu): ELU(alpha=1.0)
                (lepe): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
              )
              (2): Upsample(scale_factor=2.0, mode='nearest')
            )
            (conv): Conv2d(384, 384, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=384)
          )
          (cpe2): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=384)
          (norm2): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
          (mlp): Mlp(
            (fc1): Linear(in_features=384, out_features=1536, bias=True)
            (act): GELU(approximate='none')
            (fc2): Linear(in_features=1536, out_features=384, bias=True)
            (drop): Dropout(p=0.0, inplace=False)
          )
          (drop_path): DropPath(drop_prob=0.200)
        )
      )
      (downsample): Identity()
    )
  )
  (norm): LayerNorm((384,), eps=1e-05, elementwise_affine=True)
  (head): Linear(in_features=384, out_features=1000, bias=True)
)

EPOCH 1
 * Acc@1 0.412 Acc@5 1.708
Accuracy of the network on the 50000 test images: 0.4%
 * Acc@1 0.092 Acc@5 0.500
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 2
 * Acc@1 3.110 Acc@5 10.312
Accuracy of the network on the 50000 test images: 3.1%
 * Acc@1 0.104 Acc@5 0.498
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 3
 * Acc@1 9.568 Acc@5 24.352
Accuracy of the network on the 50000 test images: 9.6%
 * Acc@1 0.104 Acc@5 0.532
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 4
 * Acc@1 17.706 Acc@5 38.606
Accuracy of the network on the 50000 test images: 17.7%
 * Acc@1 0.100 Acc@5 0.514
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 5
 * Acc@1 25.182 Acc@5 49.382
Accuracy of the network on the 50000 test images: 25.2%
 * Acc@1 0.128 Acc@5 0.538
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 6
 * Acc@1 32.100 Acc@5 57.980
Accuracy of the network on the 50000 test images: 32.1%
 * Acc@1 0.174 Acc@5 0.520
Accuracy of the ema network on the 50000 test images: 0.2%
--
EPOCH 7
 * Acc@1 37.840 Acc@5 63.850
Accuracy of the network on the 50000 test images: 37.8%
 * Acc@1 0.174 Acc@5 0.628
Accuracy of the ema network on the 50000 test images: 0.2%
--
EPOCH 8
 * Acc@1 42.556 Acc@5 68.114
Accuracy of the network on the 50000 test images: 42.6%
 * Acc@1 0.126 Acc@5 0.874
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 9
 * Acc@1 47.154 Acc@5 72.904
Accuracy of the network on the 50000 test images: 47.2%
 * Acc@1 0.140 Acc@5 0.820
Accuracy of the ema network on the 50000 test images: 0.1%
--
EPOCH 10
 * Acc@1 50.232 Acc@5 75.610
Accuracy of the network on the 50000 test images: 50.2%
 * Acc@1 0.220 Acc@5 0.876
Accuracy of the ema network on the 50000 test images: 0.2%
--
EPOCH 11
 * Acc@1 53.298 Acc@5 78.440
Accuracy of the network on the 50000 test images: 53.3%
 * Acc@1 0.312 Acc@5 1.156
Accuracy of the ema network on the 50000 test images: 0.3%
--
EPOCH 12
 * Acc@1 55.002 Acc@5 79.402
Accuracy of the network on the 50000 test images: 55.0%
 * Acc@1 0.308 Acc@5 1.346
Accuracy of the ema network on the 50000 test images: 0.3%
--
EPOCH 13
 * Acc@1 57.534 Acc@5 81.326
Accuracy of the network on the 50000 test images: 57.5%
 * Acc@1 0.352 Acc@5 1.906
Accuracy of the ema network on the 50000 test images: 0.4%
--
EPOCH 14
 * Acc@1 58.850 Acc@5 82.558
Accuracy of the network on the 50000 test images: 58.9%
 * Acc@1 0.526 Acc@5 3.084
Accuracy of the ema network on the 50000 test images: 0.5%
--
EPOCH 15
 * Acc@1 60.512 Acc@5 84.006
Accuracy of the network on the 50000 test images: 60.5%
 * Acc@1 1.104 Acc@5 5.244
Accuracy of the ema network on the 50000 test images: 1.1%
--
EPOCH 16
 * Acc@1 61.318 Acc@5 84.554
Accuracy of the network on the 50000 test images: 61.3%
 * Acc@1 2.474 Acc@5 8.732
Accuracy of the ema network on the 50000 test images: 2.5%
--
EPOCH 17
 * Acc@1 62.456 Acc@5 85.330
Accuracy of the network on the 50000 test images: 62.5%
 * Acc@1 4.670 Acc@5 13.862
Accuracy of the ema network on the 50000 test images: 4.7%
--
EPOCH 18
 * Acc@1 63.116 Acc@5 85.796
Accuracy of the network on the 50000 test images: 63.1%
 * Acc@1 7.840 Acc@5 21.108
Accuracy of the ema network on the 50000 test images: 7.8%
--
EPOCH 19
 * Acc@1 63.686 Acc@5 86.040
Accuracy of the network on the 50000 test images: 63.7%
 * Acc@1 12.078 Acc@5 29.122
Accuracy of the ema network on the 50000 test images: 12.1%
--
EPOCH 20
 * Acc@1 64.420 Acc@5 86.622
Accuracy of the network on the 50000 test images: 64.4%
 * Acc@1 17.366 Acc@5 37.358
Accuracy of the ema network on the 50000 test images: 17.4%
--
EPOCH 21
 * Acc@1 65.496 Acc@5 87.210
Accuracy of the network on the 50000 test images: 65.5%
 * Acc@1 23.062 Acc@5 45.562
Accuracy of the ema network on the 50000 test images: 23.1%
--
EPOCH 22
 * Acc@1 66.116 Acc@5 87.798
Accuracy of the network on the 50000 test images: 66.1%
 * Acc@1 28.938 Acc@5 53.214
Accuracy of the ema network on the 50000 test images: 28.9%
--
EPOCH 23
 * Acc@1 67.238 Acc@5 88.250
Accuracy of the network on the 50000 test images: 67.2%
 * Acc@1 34.830 Acc@5 60.296
Accuracy of the ema network on the 50000 test images: 34.8%
--
EPOCH 24
 * Acc@1 67.286 Acc@5 88.386
Accuracy of the network on the 50000 test images: 67.3%
 * Acc@1 40.424 Acc@5 66.090
Accuracy of the ema network on the 50000 test images: 40.4%
--
EPOCH 25
 * Acc@1 68.334 Acc@5 89.156
Accuracy of the network on the 50000 test images: 68.3%
 * Acc@1 45.266 Acc@5 71.012
Accuracy of the ema network on the 50000 test images: 45.3%
--
EPOCH 26
 * Acc@1 68.382 Acc@5 89.302
Accuracy of the network on the 50000 test images: 68.4%
 * Acc@1 49.704 Acc@5 75.148
Accuracy of the ema network on the 50000 test images: 49.7%
--
EPOCH 27
 * Acc@1 68.794 Acc@5 89.324
Accuracy of the network on the 50000 test images: 68.8%
 * Acc@1 53.494 Acc@5 78.476
Accuracy of the ema network on the 50000 test images: 53.5%
--
EPOCH 28
 * Acc@1 69.358 Acc@5 89.824
Accuracy of the network on the 50000 test images: 69.4%
 * Acc@1 56.660 Acc@5 80.952
Accuracy of the ema network on the 50000 test images: 56.7%
--
EPOCH 29
 * Acc@1 69.340 Acc@5 89.688
Accuracy of the network on the 50000 test images: 69.3%
 * Acc@1 59.356 Acc@5 83.000
Accuracy of the ema network on the 50000 test images: 59.4%
--
EPOCH 30
 * Acc@1 70.016 Acc@5 89.966
Accuracy of the network on the 50000 test images: 70.0%
 * Acc@1 61.788 Acc@5 84.684
Accuracy of the ema network on the 50000 test images: 61.8%
--
EPOCH 31
 * Acc@1 69.828 Acc@5 90.102
Accuracy of the network on the 50000 test images: 69.8%
 * Acc@1 63.704 Acc@5 86.008
Accuracy of the ema network on the 50000 test images: 63.7%
--
EPOCH 32
 * Acc@1 70.484 Acc@5 90.344
Accuracy of the network on the 50000 test images: 70.5%
 * Acc@1 65.240 Acc@5 87.040
Accuracy of the ema network on the 50000 test images: 65.2%
--
EPOCH 33
 * Acc@1 70.382 Acc@5 90.440
Accuracy of the network on the 50000 test images: 70.4%
 * Acc@1 66.676 Acc@5 87.928
Accuracy of the ema network on the 50000 test images: 66.7%
--
EPOCH 34
 * Acc@1 70.750 Acc@5 90.660
Accuracy of the network on the 50000 test images: 70.8%
 * Acc@1 67.818 Acc@5 88.656
Accuracy of the ema network on the 50000 test images: 67.8%
--
EPOCH 35
 * Acc@1 71.254 Acc@5 90.616
Accuracy of the network on the 50000 test images: 71.3%
 * Acc@1 68.970 Acc@5 89.342
Accuracy of the ema network on the 50000 test images: 69.0%
--
EPOCH 36
 * Acc@1 71.406 Acc@5 90.640
Accuracy of the network on the 50000 test images: 71.4%
 * Acc@1 69.798 Acc@5 89.862
Accuracy of the ema network on the 50000 test images: 69.8%
--
EPOCH 37
 * Acc@1 71.916 Acc@5 91.132
Accuracy of the network on the 50000 test images: 71.9%
 * Acc@1 70.604 Acc@5 90.294
Accuracy of the ema network on the 50000 test images: 70.6%
--
EPOCH 38
 * Acc@1 71.178 Acc@5 90.460
Accuracy of the network on the 50000 test images: 71.2%
 * Acc@1 71.290 Acc@5 90.654
Accuracy of the ema network on the 50000 test images: 71.3%
--
EPOCH 39
 * Acc@1 72.028 Acc@5 91.300
Accuracy of the network on the 50000 test images: 72.0%
 * Acc@1 71.826 Acc@5 91.004
Accuracy of the ema network on the 50000 test images: 71.8%
--
EPOCH 40
 * Acc@1 72.406 Acc@5 91.404
Accuracy of the network on the 50000 test images: 72.4%
 * Acc@1 72.340 Acc@5 91.234
Accuracy of the ema network on the 50000 test images: 72.3%
--
EPOCH 41
 * Acc@1 72.498 Acc@5 91.462
Accuracy of the network on the 50000 test images: 72.5%
 * Acc@1 72.716 Acc@5 91.504
Accuracy of the ema network on the 50000 test images: 72.7%
--
EPOCH 42
 * Acc@1 72.446 Acc@5 91.400
Accuracy of the network on the 50000 test images: 72.4%
 * Acc@1 73.138 Acc@5 91.746
Accuracy of the ema network on the 50000 test images: 73.1%
--
EPOCH 43
 * Acc@1 72.592 Acc@5 91.516
Accuracy of the network on the 50000 test images: 72.6%
 * Acc@1 73.514 Acc@5 91.932
Accuracy of the ema network on the 50000 test images: 73.5%
--
EPOCH 44
 * Acc@1 72.112 Acc@5 91.154
Accuracy of the network on the 50000 test images: 72.1%
 * Acc@1 73.812 Acc@5 92.088
Accuracy of the ema network on the 50000 test images: 73.8%
--
EPOCH 45
 * Acc@1 72.992 Acc@5 91.794
Accuracy of the network on the 50000 test images: 73.0%
 * Acc@1 74.092 Acc@5 92.238
Accuracy of the ema network on the 50000 test images: 74.1%
--
EPOCH 46
 * Acc@1 72.982 Acc@5 91.664
Accuracy of the network on the 50000 test images: 73.0%
 * Acc@1 74.420 Acc@5 92.338
Accuracy of the ema network on the 50000 test images: 74.4%
--
EPOCH 47
 * Acc@1 73.250 Acc@5 91.838
Accuracy of the network on the 50000 test images: 73.2%
 * Acc@1 74.680 Acc@5 92.446
Accuracy of the ema network on the 50000 test images: 74.7%
--
EPOCH 48
 * Acc@1 73.188 Acc@5 91.824
Accuracy of the network on the 50000 test images: 73.2%
 * Acc@1 74.898 Acc@5 92.620
Accuracy of the ema network on the 50000 test images: 74.9%
--
EPOCH 49
 * Acc@1 73.186 Acc@5 91.754
Accuracy of the network on the 50000 test images: 73.2%
 * Acc@1 75.124 Acc@5 92.756
Accuracy of the ema network on the 50000 test images: 75.1%
--
EPOCH 50
 * Acc@1 73.152 Acc@5 91.780
Accuracy of the network on the 50000 test images: 73.2%
 * Acc@1 75.312 Acc@5 92.858
Accuracy of the ema network on the 50000 test images: 75.3%
--
EPOCH 51
 * Acc@1 73.458 Acc@5 91.942
Accuracy of the network on the 50000 test images: 73.5%
 * Acc@1 75.430 Acc@5 92.920
Accuracy of the ema network on the 50000 test images: 75.4%
--
EPOCH 52
 * Acc@1 73.250 Acc@5 92.064
Accuracy of the network on the 50000 test images: 73.3%
 * Acc@1 75.572 Acc@5 93.002
Accuracy of the ema network on the 50000 test images: 75.6%
--
EPOCH 53
 * Acc@1 73.554 Acc@5 92.070
Accuracy of the network on the 50000 test images: 73.6%
 * Acc@1 75.704 Acc@5 93.036
Accuracy of the ema network on the 50000 test images: 75.7%
--
EPOCH 54
 * Acc@1 73.392 Acc@5 91.924
Accuracy of the network on the 50000 test images: 73.4%
 * Acc@1 75.876 Acc@5 93.164
Accuracy of the ema network on the 50000 test images: 75.9%
--
EPOCH 55
 * Acc@1 73.582 Acc@5 92.154
Accuracy of the network on the 50000 test images: 73.6%
 * Acc@1 76.012 Acc@5 93.240
Accuracy of the ema network on the 50000 test images: 76.0%
--
EPOCH 56
 * Acc@1 73.716 Acc@5 92.216
Accuracy of the network on the 50000 test images: 73.7%
 * Acc@1 76.134 Acc@5 93.288
Accuracy of the ema network on the 50000 test images: 76.1%
--
EPOCH 57
 * Acc@1 73.818 Acc@5 92.218
Accuracy of the network on the 50000 test images: 73.8%
 * Acc@1 76.172 Acc@5 93.334
Accuracy of the ema network on the 50000 test images: 76.2%
--
EPOCH 58
 * Acc@1 74.080 Acc@5 92.328
Accuracy of the network on the 50000 test images: 74.1%
 * Acc@1 76.286 Acc@5 93.408
Accuracy of the ema network on the 50000 test images: 76.3%
--
EPOCH 59
 * Acc@1 73.740 Acc@5 92.346
Accuracy of the network on the 50000 test images: 73.7%
 * Acc@1 76.398 Acc@5 93.456
Accuracy of the ema network on the 50000 test images: 76.4%
--
EPOCH 60
 * Acc@1 74.038 Acc@5 92.252
Accuracy of the network on the 50000 test images: 74.0%
 * Acc@1 76.518 Acc@5 93.506
Accuracy of the ema network on the 50000 test images: 76.5%
--
EPOCH 61
 * Acc@1 74.384 Acc@5 92.460
Accuracy of the network on the 50000 test images: 74.4%
 * Acc@1 76.610 Acc@5 93.564
Accuracy of the ema network on the 50000 test images: 76.6%
--
EPOCH 62
 * Acc@1 74.270 Acc@5 92.532
Accuracy of the network on the 50000 test images: 74.3%
 * Acc@1 76.626 Acc@5 93.590
Accuracy of the ema network on the 50000 test images: 76.6%
--
EPOCH 63
 * Acc@1 73.992 Acc@5 92.480
Accuracy of the network on the 50000 test images: 74.0%
 * Acc@1 76.740 Acc@5 93.664
Accuracy of the ema network on the 50000 test images: 76.7%
--
EPOCH 64
 * Acc@1 74.452 Acc@5 92.650
Accuracy of the network on the 50000 test images: 74.5%
 * Acc@1 76.762 Acc@5 93.688
Accuracy of the ema network on the 50000 test images: 76.8%
--
EPOCH 65
 * Acc@1 74.430 Acc@5 92.348
Accuracy of the network on the 50000 test images: 74.4%
 * Acc@1 76.876 Acc@5 93.732
Accuracy of the ema network on the 50000 test images: 76.9%
--
EPOCH 66
 * Acc@1 74.318 Acc@5 92.582
Accuracy of the network on the 50000 test images: 74.3%
 * Acc@1 77.022 Acc@5 93.806
Accuracy of the ema network on the 50000 test images: 77.0%
--
EPOCH 67
 * Acc@1 74.564 Acc@5 92.676
Accuracy of the network on the 50000 test images: 74.6%
 * Acc@1 77.076 Acc@5 93.830
Accuracy of the ema network on the 50000 test images: 77.1%
--
EPOCH 68
 * Acc@1 74.570 Acc@5 92.498
Accuracy of the network on the 50000 test images: 74.6%
 * Acc@1 77.166 Acc@5 93.864
Accuracy of the ema network on the 50000 test images: 77.2%
--
EPOCH 69
 * Acc@1 74.372 Acc@5 92.518
Accuracy of the network on the 50000 test images: 74.4%
 * Acc@1 77.218 Acc@5 93.886
Accuracy of the ema network on the 50000 test images: 77.2%
--
EPOCH 70
 * Acc@1 74.460 Acc@5 92.562
Accuracy of the network on the 50000 test images: 74.5%
 * Acc@1 77.266 Acc@5 93.908
Accuracy of the ema network on the 50000 test images: 77.3%
--
EPOCH 71
 * Acc@1 74.516 Acc@5 92.604
Accuracy of the network on the 50000 test images: 74.5%
 * Acc@1 77.342 Acc@5 93.944
Accuracy of the ema network on the 50000 test images: 77.3%
--
EPOCH 72
 * Acc@1 74.838 Acc@5 92.754
Accuracy of the network on the 50000 test images: 74.8%
 * Acc@1 77.394 Acc@5 93.956
Accuracy of the ema network on the 50000 test images: 77.4%
--
EPOCH 73
 * Acc@1 74.670 Acc@5 92.712
Accuracy of the network on the 50000 test images: 74.7%
 * Acc@1 77.490 Acc@5 94.014
Accuracy of the ema network on the 50000 test images: 77.5%
--
EPOCH 74
 * Acc@1 74.650 Acc@5 92.560
Accuracy of the network on the 50000 test images: 74.7%
 * Acc@1 77.584 Acc@5 94.030
Accuracy of the ema network on the 50000 test images: 77.6%
--
EPOCH 75
 * Acc@1 74.876 Acc@5 92.852
Accuracy of the network on the 50000 test images: 74.9%
 * Acc@1 77.664 Acc@5 94.064
Accuracy of the ema network on the 50000 test images: 77.7%
--
EPOCH 76
 * Acc@1 75.616 Acc@5 93.050
Accuracy of the network on the 50000 test images: 75.6%
 * Acc@1 77.656 Acc@5 94.048
Accuracy of the ema network on the 50000 test images: 77.7%
--
EPOCH 77
 * Acc@1 75.930 Acc@5 93.186
Accuracy of the network on the 50000 test images: 75.9%
 * Acc@1 77.704 Acc@5 94.044
Accuracy of the ema network on the 50000 test images: 77.7%
--
EPOCH 78
 * Acc@1 75.788 Acc@5 93.246
Accuracy of the network on the 50000 test images: 75.8%
 * Acc@1 77.718 Acc@5 94.084
Accuracy of the ema network on the 50000 test images: 77.7%
--
EPOCH 79
 * Acc@1 75.824 Acc@5 93.166
Accuracy of the network on the 50000 test images: 75.8%
 * Acc@1 77.746 Acc@5 94.058
Accuracy of the ema network on the 50000 test images: 77.7%
--
EPOCH 80
 * Acc@1 75.994 Acc@5 93.250
Accuracy of the network on the 50000 test images: 76.0%
 * Acc@1 77.750 Acc@5 94.092
Accuracy of the ema network on the 50000 test images: 77.8%
--
EPOCH 81
 * Acc@1 75.824 Acc@5 93.154
Accuracy of the network on the 50000 test images: 75.8%
 * Acc@1 77.810 Acc@5 94.112
Accuracy of the ema network on the 50000 test images: 77.8%
--
EPOCH 82
 * Acc@1 76.218 Acc@5 93.348
Accuracy of the network on the 50000 test images: 76.2%
 * Acc@1 77.834 Acc@5 94.120
Accuracy of the ema network on the 50000 test images: 77.8%
--
EPOCH 83
 * Acc@1 76.132 Acc@5 93.308
Accuracy of the network on the 50000 test images: 76.1%
 * Acc@1 77.852 Acc@5 94.126
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 84
 * Acc@1 76.030 Acc@5 93.338
Accuracy of the network on the 50000 test images: 76.0%
 * Acc@1 77.870 Acc@5 94.158
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 85
 * Acc@1 76.198 Acc@5 93.332
Accuracy of the network on the 50000 test images: 76.2%
 * Acc@1 77.874 Acc@5 94.136
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 86
 * Acc@1 76.222 Acc@5 93.356
Accuracy of the network on the 50000 test images: 76.2%
 * Acc@1 77.864 Acc@5 94.156
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 87
 * Acc@1 76.308 Acc@5 93.456
Accuracy of the network on the 50000 test images: 76.3%
 * Acc@1 77.862 Acc@5 94.184
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 88
 * Acc@1 76.470 Acc@5 93.510
Accuracy of the network on the 50000 test images: 76.5%
 * Acc@1 77.902 Acc@5 94.178
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 89
 * Acc@1 76.462 Acc@5 93.452
Accuracy of the network on the 50000 test images: 76.5%
 * Acc@1 77.900 Acc@5 94.202
Accuracy of the ema network on the 50000 test images: 77.9%
--
EPOCH 90
 * Acc@1 76.670 Acc@5 93.556
Accuracy of the network on the 50000 test images: 76.7%
 * Acc@1 77.954 Acc@5 94.194
Accuracy of the ema network on the 50000 test images: 78.0%
--
EPOCH 91
 * Acc@1 76.292 Acc@5 93.506
Accuracy of the network on the 50000 test images: 76.3%
 * Acc@1 77.984 Acc@5 94.210
Accuracy of the ema network on the 50000 test images: 78.0%
--
EPOCH 92
 * Acc@1 76.462 Acc@5 93.604
Accuracy of the network on the 50000 test images: 76.5%
 * Acc@1 77.984 Acc@5 94.236
Accuracy of the ema network on the 50000 test images: 78.0%
--
EPOCH 93
 * Acc@1 76.282 Acc@5 93.516
Accuracy of the network on the 50000 test images: 76.3%
 * Acc@1 78.058 Acc@5 94.272
Accuracy of the ema network on the 50000 test images: 78.1%
--
EPOCH 94
 * Acc@1 76.508 Acc@5 93.448
Accuracy of the network on the 50000 test images: 76.5%
 * Acc@1 78.084 Acc@5 94.290
Accuracy of the ema network on the 50000 test images: 78.1%
--
EPOCH 95
 * Acc@1 76.408 Acc@5 93.542
Accuracy of the network on the 50000 test images: 76.4%
 * Acc@1 78.130 Acc@5 94.320
Accuracy of the ema network on the 50000 test images: 78.1%
--
EPOCH 96
 * Acc@1 76.680 Acc@5 93.728
Accuracy of the network on the 50000 test images: 76.7%
 * Acc@1 78.178 Acc@5 94.328
Accuracy of the ema network on the 50000 test images: 78.2%
--
EPOCH 97
 * Acc@1 76.426 Acc@5 93.518
Accuracy of the network on the 50000 test images: 76.4%
 * Acc@1 78.216 Acc@5 94.330
Accuracy of the ema network on the 50000 test images: 78.2%
--
EPOCH 98
 * Acc@1 76.396 Acc@5 93.736
Accuracy of the network on the 50000 test images: 76.4%
 * Acc@1 78.264 Acc@5 94.338
Accuracy of the ema network on the 50000 test images: 78.3%
--
EPOCH 99
 * Acc@1 76.570 Acc@5 93.772
Accuracy of the network on the 50000 test images: 76.6%
 * Acc@1 78.256 Acc@5 94.348
Accuracy of the ema network on the 50000 test images: 78.3%
--
EPOCH 100
 * Acc@1 76.770 Acc@5 93.726
Accuracy of the network on the 50000 test images: 76.8%
 * Acc@1 78.276 Acc@5 94.392
Accuracy of the ema network on the 50000 test images: 78.3%
--
EPOCH 101
 * Acc@1 76.566 Acc@5 93.616
Accuracy of the network on the 50000 test images: 76.6%
 * Acc@1 78.254 Acc@5 94.392
Accuracy of the ema network on the 50000 test images: 78.3%
--
EPOCH 102
 * Acc@1 76.944 Acc@5 93.654
Accuracy of the network on the 50000 test images: 76.9%
 * Acc@1 78.306 Acc@5 94.420
Accuracy of the ema network on the 50000 test images: 78.3%
--
EPOCH 103
 * Acc@1 76.478 Acc@5 93.664
Accuracy of the network on the 50000 test images: 76.5%
 * Acc@1 78.364 Acc@5 94.428
Accuracy of the ema network on the 50000 test images: 78.4%
--
EPOCH 104
 * Acc@1 76.832 Acc@5 93.728
Accuracy of the network on the 50000 test images: 76.8%
 * Acc@1 78.352 Acc@5 94.430
Accuracy of the ema network on the 50000 test images: 78.4%
--
EPOCH 105
 * Acc@1 76.826 Acc@5 93.750
Accuracy of the network on the 50000 test images: 76.8%
 * Acc@1 78.388 Acc@5 94.478
Accuracy of the ema network on the 50000 test images: 78.4%
--
EPOCH 106
 * Acc@1 76.804 Acc@5 93.742
Accuracy of the network on the 50000 test images: 76.8%
 * Acc@1 78.400 Acc@5 94.486
Accuracy of the ema network on the 50000 test images: 78.4%
--
EPOCH 107
 * Acc@1 77.120 Acc@5 93.914
Accuracy of the network on the 50000 test images: 77.1%
 * Acc@1 78.416 Acc@5 94.504
Accuracy of the ema network on the 50000 test images: 78.4%
--
EPOCH 108
 * Acc@1 77.118 Acc@5 93.822
Accuracy of the network on the 50000 test images: 77.1%
 * Acc@1 78.430 Acc@5 94.556
Accuracy of the ema network on the 50000 test images: 78.4%
--
EPOCH 109
 * Acc@1 77.002 Acc@5 93.768
Accuracy of the network on the 50000 test images: 77.0%
 * Acc@1 78.472 Acc@5 94.566
Accuracy of the ema network on the 50000 test images: 78.5%
--
EPOCH 110
 * Acc@1 77.040 Acc@5 93.828
Accuracy of the network on the 50000 test images: 77.0%
 * Acc@1 78.488 Acc@5 94.572
Accuracy of the ema network on the 50000 test images: 78.5%
--
EPOCH 111
 * Acc@1 77.084 Acc@5 93.822
Accuracy of the network on the 50000 test images: 77.1%
 * Acc@1 78.552 Acc@5 94.596
Accuracy of the ema network on the 50000 test images: 78.6%
--
EPOCH 112
 * Acc@1 77.306 Acc@5 93.926
Accuracy of the network on the 50000 test images: 77.3%
 * Acc@1 78.570 Acc@5 94.592
Accuracy of the ema network on the 50000 test images: 78.6%
--
EPOCH 113
 * Acc@1 77.174 Acc@5 93.962
Accuracy of the network on the 50000 test images: 77.2%
 * Acc@1 78.618 Acc@5 94.574
Accuracy of the ema network on the 50000 test images: 78.6%
--
EPOCH 114
 * Acc@1 77.284 Acc@5 93.982
Accuracy of the network on the 50000 test images: 77.3%
 * Acc@1 78.658 Acc@5 94.590
Accuracy of the ema network on the 50000 test images: 78.7%
--
EPOCH 115
 * Acc@1 76.966 Acc@5 93.870
Accuracy of the network on the 50000 test images: 77.0%
 * Acc@1 78.690 Acc@5 94.604
Accuracy of the ema network on the 50000 test images: 78.7%
--
EPOCH 116
 * Acc@1 77.348 Acc@5 93.968
Accuracy of the network on the 50000 test images: 77.3%
 * Acc@1 78.708 Acc@5 94.608
Accuracy of the ema network on the 50000 test images: 78.7%
--
EPOCH 117
 * Acc@1 77.338 Acc@5 93.968
Accuracy of the network on the 50000 test images: 77.3%
 * Acc@1 78.750 Acc@5 94.620
Accuracy of the ema network on the 50000 test images: 78.8%
--
EPOCH 118
 * Acc@1 77.204 Acc@5 93.834
Accuracy of the network on the 50000 test images: 77.2%
 * Acc@1 78.774 Acc@5 94.618
Accuracy of the ema network on the 50000 test images: 78.8%
--
EPOCH 119
 * Acc@1 77.512 Acc@5 93.996
Accuracy of the network on the 50000 test images: 77.5%
 * Acc@1 78.822 Acc@5 94.636
Accuracy of the ema network on the 50000 test images: 78.8%
--
EPOCH 120
 * Acc@1 77.286 Acc@5 93.816
Accuracy of the network on the 50000 test images: 77.3%
 * Acc@1 78.850 Acc@5 94.650
Accuracy of the ema network on the 50000 test images: 78.9%
--
EPOCH 121
 * Acc@1 77.466 Acc@5 93.998
Accuracy of the network on the 50000 test images: 77.5%
 * Acc@1 78.878 Acc@5 94.662
Accuracy of the ema network on the 50000 test images: 78.9%
--
EPOCH 122
 * Acc@1 77.500 Acc@5 93.966
Accuracy of the network on the 50000 test images: 77.5%
 * Acc@1 78.840 Acc@5 94.706
Accuracy of the ema network on the 50000 test images: 78.8%
--
EPOCH 123
 * Acc@1 77.542 Acc@5 94.058
Accuracy of the network on the 50000 test images: 77.5%
 * Acc@1 78.876 Acc@5 94.714
Accuracy of the ema network on the 50000 test images: 78.9%
--
EPOCH 124
 * Acc@1 77.554 Acc@5 94.060
Accuracy of the network on the 50000 test images: 77.6%
 * Acc@1 78.918 Acc@5 94.708
Accuracy of the ema network on the 50000 test images: 78.9%
--
EPOCH 125
 * Acc@1 77.520 Acc@5 94.022
Accuracy of the network on the 50000 test images: 77.5%
 * Acc@1 78.924 Acc@5 94.710
Accuracy of the ema network on the 50000 test images: 78.9%
--
EPOCH 126
 * Acc@1 77.628 Acc@5 94.094
Accuracy of the network on the 50000 test images: 77.6%
 * Acc@1 78.984 Acc@5 94.716
Accuracy of the ema network on the 50000 test images: 79.0%
--
EPOCH 127
 * Acc@1 77.376 Acc@5 94.130
Accuracy of the network on the 50000 test images: 77.4%
 * Acc@1 79.006 Acc@5 94.720
Accuracy of the ema network on the 50000 test images: 79.0%
--
EPOCH 128
 * Acc@1 77.592 Acc@5 94.174
Accuracy of the network on the 50000 test images: 77.6%
 * Acc@1 79.068 Acc@5 94.736
Accuracy of the ema network on the 50000 test images: 79.1%
--
EPOCH 129
 * Acc@1 77.446 Acc@5 94.076
Accuracy of the network on the 50000 test images: 77.4%
 * Acc@1 79.072 Acc@5 94.764
Accuracy of the ema network on the 50000 test images: 79.1%
--
EPOCH 130
 * Acc@1 77.404 Acc@5 94.140
Accuracy of the network on the 50000 test images: 77.4%
 * Acc@1 79.082 Acc@5 94.770
Accuracy of the ema network on the 50000 test images: 79.1%
--
EPOCH 131
 * Acc@1 77.676 Acc@5 94.264
Accuracy of the network on the 50000 test images: 77.7%
 * Acc@1 79.110 Acc@5 94.784
Accuracy of the ema network on the 50000 test images: 79.1%
--
EPOCH 132
 * Acc@1 77.824 Acc@5 94.320
Accuracy of the network on the 50000 test images: 77.8%
 * Acc@1 79.134 Acc@5 94.792
Accuracy of the ema network on the 50000 test images: 79.1%
--
EPOCH 133
 * Acc@1 77.794 Acc@5 94.214
Accuracy of the network on the 50000 test images: 77.8%
 * Acc@1 79.156 Acc@5 94.796
Accuracy of the ema network on the 50000 test images: 79.2%
--
EPOCH 134
 * Acc@1 77.798 Acc@5 94.206
Accuracy of the network on the 50000 test images: 77.8%
 * Acc@1 79.188 Acc@5 94.832
Accuracy of the ema network on the 50000 test images: 79.2%
--
EPOCH 135
 * Acc@1 77.882 Acc@5 94.282
Accuracy of the network on the 50000 test images: 77.9%
 * Acc@1 79.174 Acc@5 94.874
Accuracy of the ema network on the 50000 test images: 79.2%
--
EPOCH 136
 * Acc@1 78.184 Acc@5 94.366
Accuracy of the network on the 50000 test images: 78.2%
 * Acc@1 79.180 Acc@5 94.882
Accuracy of the ema network on the 50000 test images: 79.2%
--
EPOCH 137
 * Acc@1 77.898 Acc@5 94.306
Accuracy of the network on the 50000 test images: 77.9%
 * Acc@1 79.232 Acc@5 94.922
Accuracy of the ema network on the 50000 test images: 79.2%
--
EPOCH 138
 * Acc@1 77.984 Acc@5 94.214
Accuracy of the network on the 50000 test images: 78.0%
 * Acc@1 79.212 Acc@5 94.924
Accuracy of the ema network on the 50000 test images: 79.2%
--
EPOCH 139
 * Acc@1 77.902 Acc@5 94.310
Accuracy of the network on the 50000 test images: 77.9%
 * Acc@1 79.308 Acc@5 94.954
Accuracy of the ema network on the 50000 test images: 79.3%
--
EPOCH 140
 * Acc@1 77.968 Acc@5 94.320
Accuracy of the network on the 50000 test images: 78.0%
 * Acc@1 79.346 Acc@5 94.972
Accuracy of the ema network on the 50000 test images: 79.3%
--
EPOCH 141
 * Acc@1 77.874 Acc@5 94.250
Accuracy of the network on the 50000 test images: 77.9%
 * Acc@1 79.370 Acc@5 94.982
Accuracy of the ema network on the 50000 test images: 79.4%
--
EPOCH 142
 * Acc@1 78.024 Acc@5 94.418
Accuracy of the network on the 50000 test images: 78.0%
 * Acc@1 79.394 Acc@5 94.988
Accuracy of the ema network on the 50000 test images: 79.4%
--
EPOCH 143
 * Acc@1 78.040 Acc@5 94.364
Accuracy of the network on the 50000 test images: 78.0%
 * Acc@1 79.382 Acc@5 94.986
Accuracy of the ema network on the 50000 test images: 79.4%
--
EPOCH 144
 * Acc@1 78.012 Acc@5 94.424
Accuracy of the network on the 50000 test images: 78.0%
 * Acc@1 79.364 Acc@5 94.998
Accuracy of the ema network on the 50000 test images: 79.4%
--
EPOCH 145
 * Acc@1 78.202 Acc@5 94.322
Accuracy of the network on the 50000 test images: 78.2%
 * Acc@1 79.386 Acc@5 95.008
Accuracy of the ema network on the 50000 test images: 79.4%
--
EPOCH 146
 * Acc@1 78.152 Acc@5 94.442
Accuracy of the network on the 50000 test images: 78.2%
 * Acc@1 79.412 Acc@5 95.020
Accuracy of the ema network on the 50000 test images: 79.4%
--
EPOCH 147
 * Acc@1 78.080 Acc@5 94.452
Accuracy of the network on the 50000 test images: 78.1%
 * Acc@1 79.496 Acc@5 95.044
Accuracy of the ema network on the 50000 test images: 79.5%
--
EPOCH 148
 * Acc@1 78.248 Acc@5 94.584
Accuracy of the network on the 50000 test images: 78.2%
 * Acc@1 79.518 Acc@5 95.044
Accuracy of the ema network on the 50000 test images: 79.5%
--
EPOCH 149
 * Acc@1 78.282 Acc@5 94.500
Accuracy of the network on the 50000 test images: 78.3%
 * Acc@1 79.532 Acc@5 95.046
Accuracy of the ema network on the 50000 test images: 79.5%
--
EPOCH 150
 * Acc@1 78.242 Acc@5 94.444
Accuracy of the network on the 50000 test images: 78.2%
 * Acc@1 79.578 Acc@5 95.076
Accuracy of the ema network on the 50000 test images: 79.6%
--
EPOCH 151
 * Acc@1 78.332 Acc@5 94.534
Accuracy of the network on the 50000 test images: 78.3%
 * Acc@1 79.596 Acc@5 95.090
Accuracy of the ema network on the 50000 test images: 79.6%
--
EPOCH 152
 * Acc@1 78.402 Acc@5 94.538
Accuracy of the network on the 50000 test images: 78.4%
 * Acc@1 79.656 Acc@5 95.104
Accuracy of the ema network on the 50000 test images: 79.7%
--
EPOCH 153
 * Acc@1 78.352 Acc@5 94.544
Accuracy of the network on the 50000 test images: 78.4%
 * Acc@1 79.694 Acc@5 95.100
Accuracy of the ema network on the 50000 test images: 79.7%
--
EPOCH 154
 * Acc@1 78.514 Acc@5 94.574
Accuracy of the network on the 50000 test images: 78.5%
 * Acc@1 79.716 Acc@5 95.088
Accuracy of the ema network on the 50000 test images: 79.7%
--
EPOCH 155
 * Acc@1 78.412 Acc@5 94.552
Accuracy of the network on the 50000 test images: 78.4%
 * Acc@1 79.728 Acc@5 95.128
Accuracy of the ema network on the 50000 test images: 79.7%
--
EPOCH 156
 * Acc@1 78.520 Acc@5 94.646
Accuracy of the network on the 50000 test images: 78.5%
 * Acc@1 79.708 Acc@5 95.132
Accuracy of the ema network on the 50000 test images: 79.7%
--
EPOCH 157
 * Acc@1 78.678 Acc@5 94.636
Accuracy of the network on the 50000 test images: 78.7%
 * Acc@1 79.768 Acc@5 95.144
Accuracy of the ema network on the 50000 test images: 79.8%
--
EPOCH 158
 * Acc@1 78.548 Acc@5 94.688
Accuracy of the network on the 50000 test images: 78.5%
 * Acc@1 79.802 Acc@5 95.184
Accuracy of the ema network on the 50000 test images: 79.8%
--
EPOCH 159
 * Acc@1 78.668 Acc@5 94.664
Accuracy of the network on the 50000 test images: 78.7%
 * Acc@1 79.796 Acc@5 95.202
Accuracy of the ema network on the 50000 test images: 79.8%
--
EPOCH 160
 * Acc@1 78.588 Acc@5 94.512
Accuracy of the network on the 50000 test images: 78.6%
 * Acc@1 79.772 Acc@5 95.204
Accuracy of the ema network on the 50000 test images: 79.8%
--
EPOCH 161
 * Acc@1 78.626 Acc@5 94.554
Accuracy of the network on the 50000 test images: 78.6%
 * Acc@1 79.798 Acc@5 95.202
Accuracy of the ema network on the 50000 test images: 79.8%
--
EPOCH 162
 * Acc@1 78.600 Acc@5 94.598
Accuracy of the network on the 50000 test images: 78.6%
 * Acc@1 79.828 Acc@5 95.218
Accuracy of the ema network on the 50000 test images: 79.8%
--
EPOCH 163
 * Acc@1 78.674 Acc@5 94.738
Accuracy of the network on the 50000 test images: 78.7%
 * Acc@1 79.850 Acc@5 95.228
Accuracy of the ema network on the 50000 test images: 79.9%
--
EPOCH 164
 * Acc@1 78.772 Acc@5 94.708
Accuracy of the network on the 50000 test images: 78.8%
 * Acc@1 79.892 Acc@5 95.240
Accuracy of the ema network on the 50000 test images: 79.9%
--
EPOCH 165
 * Acc@1 78.946 Acc@5 94.726
Accuracy of the network on the 50000 test images: 78.9%
 * Acc@1 79.930 Acc@5 95.278
Accuracy of the ema network on the 50000 test images: 79.9%
--
EPOCH 166
 * Acc@1 78.808 Acc@5 94.760
Accuracy of the network on the 50000 test images: 78.8%
 * Acc@1 79.978 Acc@5 95.256
Accuracy of the ema network on the 50000 test images: 80.0%
--
EPOCH 167
 * Acc@1 78.858 Acc@5 94.692
Accuracy of the network on the 50000 test images: 78.9%
 * Acc@1 79.986 Acc@5 95.304
Accuracy of the ema network on the 50000 test images: 80.0%
--
EPOCH 168
 * Acc@1 78.972 Acc@5 94.796
Accuracy of the network on the 50000 test images: 79.0%
 * Acc@1 79.980 Acc@5 95.312
Accuracy of the ema network on the 50000 test images: 80.0%
--
EPOCH 169
 * Acc@1 78.838 Acc@5 94.732
Accuracy of the network on the 50000 test images: 78.8%
 * Acc@1 80.008 Acc@5 95.334
Accuracy of the ema network on the 50000 test images: 80.0%
--
EPOCH 170
 * Acc@1 79.100 Acc@5 94.800
Accuracy of the network on the 50000 test images: 79.1%
 * Acc@1 80.076 Acc@5 95.356
Accuracy of the ema network on the 50000 test images: 80.1%
--
EPOCH 171
 * Acc@1 79.178 Acc@5 94.836
Accuracy of the network on the 50000 test images: 79.2%
 * Acc@1 80.090 Acc@5 95.356
Accuracy of the ema network on the 50000 test images: 80.1%
--
EPOCH 172
 * Acc@1 79.064 Acc@5 94.894
Accuracy of the network on the 50000 test images: 79.1%
 * Acc@1 80.098 Acc@5 95.360
Accuracy of the ema network on the 50000 test images: 80.1%
--
EPOCH 173
 * Acc@1 79.064 Acc@5 94.898
Accuracy of the network on the 50000 test images: 79.1%
 * Acc@1 80.146 Acc@5 95.376
Accuracy of the ema network on the 50000 test images: 80.1%
--
EPOCH 174
 * Acc@1 79.168 Acc@5 94.768
Accuracy of the network on the 50000 test images: 79.2%
 * Acc@1 80.176 Acc@5 95.390
Accuracy of the ema network on the 50000 test images: 80.2%
--
EPOCH 175
 * Acc@1 78.982 Acc@5 94.888
Accuracy of the network on the 50000 test images: 79.0%
 * Acc@1 80.220 Acc@5 95.390
Accuracy of the ema network on the 50000 test images: 80.2%
--
EPOCH 176
 * Acc@1 79.280 Acc@5 94.864
Accuracy of the network on the 50000 test images: 79.3%
 * Acc@1 80.222 Acc@5 95.402
Accuracy of the ema network on the 50000 test images: 80.2%
--
EPOCH 177
 * Acc@1 79.162 Acc@5 94.952
Accuracy of the network on the 50000 test images: 79.2%
 * Acc@1 80.276 Acc@5 95.416
Accuracy of the ema network on the 50000 test images: 80.3%
--
EPOCH 178
 * Acc@1 79.264 Acc@5 94.906
Accuracy of the network on the 50000 test images: 79.3%
 * Acc@1 80.282 Acc@5 95.406
Accuracy of the ema network on the 50000 test images: 80.3%
--
EPOCH 179
 * Acc@1 79.280 Acc@5 94.914
Accuracy of the network on the 50000 test images: 79.3%
 * Acc@1 80.296 Acc@5 95.416
Accuracy of the ema network on the 50000 test images: 80.3%
--
EPOCH 180
 * Acc@1 79.412 Acc@5 95.014
Accuracy of the network on the 50000 test images: 79.4%
 * Acc@1 80.304 Acc@5 95.416
Accuracy of the ema network on the 50000 test images: 80.3%
--
EPOCH 181
 * Acc@1 79.340 Acc@5 94.968
Accuracy of the network on the 50000 test images: 79.3%
 * Acc@1 80.352 Acc@5 95.420
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 182
 * Acc@1 79.360 Acc@5 94.948
Accuracy of the network on the 50000 test images: 79.4%
 * Acc@1 80.390 Acc@5 95.446
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 183
 * Acc@1 79.446 Acc@5 95.018
Accuracy of the network on the 50000 test images: 79.4%
 * Acc@1 80.394 Acc@5 95.472
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 184
 * Acc@1 79.436 Acc@5 95.016
Accuracy of the network on the 50000 test images: 79.4%
 * Acc@1 80.358 Acc@5 95.464
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 185
 * Acc@1 79.460 Acc@5 95.036
Accuracy of the network on the 50000 test images: 79.5%
 * Acc@1 80.392 Acc@5 95.478
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 186
 * Acc@1 79.576 Acc@5 95.016
Accuracy of the network on the 50000 test images: 79.6%
 * Acc@1 80.396 Acc@5 95.492
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 187
 * Acc@1 79.524 Acc@5 95.102
Accuracy of the network on the 50000 test images: 79.5%
 * Acc@1 80.438 Acc@5 95.482
Accuracy of the ema network on the 50000 test images: 80.4%
--
EPOCH 188
 * Acc@1 79.588 Acc@5 95.064
Accuracy of the network on the 50000 test images: 79.6%
 * Acc@1 80.462 Acc@5 95.494
Accuracy of the ema network on the 50000 test images: 80.5%
--
EPOCH 189
 * Acc@1 79.438 Acc@5 94.986
Accuracy of the network on the 50000 test images: 79.4%
 * Acc@1 80.514 Acc@5 95.484
Accuracy of the ema network on the 50000 test images: 80.5%
--
EPOCH 190
 * Acc@1 79.596 Acc@5 95.126
Accuracy of the network on the 50000 test images: 79.6%
 * Acc@1 80.538 Acc@5 95.480
Accuracy of the ema network on the 50000 test images: 80.5%
--
EPOCH 191
 * Acc@1 79.774 Acc@5 95.182
Accuracy of the network on the 50000 test images: 79.8%
 * Acc@1 80.552 Acc@5 95.502
Accuracy of the ema network on the 50000 test images: 80.6%
--
EPOCH 192
 * Acc@1 79.734 Acc@5 95.144
Accuracy of the network on the 50000 test images: 79.7%
 * Acc@1 80.582 Acc@5 95.518
Accuracy of the ema network on the 50000 test images: 80.6%
--
EPOCH 193
 * Acc@1 79.614 Acc@5 95.126
Accuracy of the network on the 50000 test images: 79.6%
 * Acc@1 80.588 Acc@5 95.520
Accuracy of the ema network on the 50000 test images: 80.6%
--
EPOCH 194
 * Acc@1 79.860 Acc@5 95.248
Accuracy of the network on the 50000 test images: 79.9%
 * Acc@1 80.614 Acc@5 95.514
Accuracy of the ema network on the 50000 test images: 80.6%
--
EPOCH 195
 * Acc@1 79.882 Acc@5 95.226
Accuracy of the network on the 50000 test images: 79.9%
 * Acc@1 80.638 Acc@5 95.546
Accuracy of the ema network on the 50000 test images: 80.6%
--
EPOCH 196
 * Acc@1 79.714 Acc@5 95.114
Accuracy of the network on the 50000 test images: 79.7%
 * Acc@1 80.668 Acc@5 95.562
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 197
 * Acc@1 79.802 Acc@5 95.286
Accuracy of the network on the 50000 test images: 79.8%
 * Acc@1 80.720 Acc@5 95.566
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 198
 * Acc@1 79.964 Acc@5 95.174
Accuracy of the network on the 50000 test images: 80.0%
 * Acc@1 80.726 Acc@5 95.586
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 199
 * Acc@1 79.952 Acc@5 95.298
Accuracy of the network on the 50000 test images: 80.0%
 * Acc@1 80.744 Acc@5 95.608
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 200
 * Acc@1 79.998 Acc@5 95.280
Accuracy of the network on the 50000 test images: 80.0%
 * Acc@1 80.740 Acc@5 95.626
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 201
 * Acc@1 79.838 Acc@5 95.280
Accuracy of the network on the 50000 test images: 79.8%
 * Acc@1 80.752 Acc@5 95.638
Accuracy of the ema network on the 50000 test images: 80.8%
--
EPOCH 202
 * Acc@1 79.926 Acc@5 95.204
Accuracy of the network on the 50000 test images: 79.9%
 * Acc@1 80.726 Acc@5 95.664
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 203
 * Acc@1 79.910 Acc@5 95.236
Accuracy of the network on the 50000 test images: 79.9%
 * Acc@1 80.770 Acc@5 95.660
Accuracy of the ema network on the 50000 test images: 80.8%
--
EPOCH 204
 * Acc@1 80.078 Acc@5 95.344
Accuracy of the network on the 50000 test images: 80.1%
 * Acc@1 80.728 Acc@5 95.646
Accuracy of the ema network on the 50000 test images: 80.7%
--
EPOCH 205
 * Acc@1 80.068 Acc@5 95.270
Accuracy of the network on the 50000 test images: 80.1%
 * Acc@1 80.790 Acc@5 95.652
Accuracy of the ema network on the 50000 test images: 80.8%
--
EPOCH 206
 * Acc@1 80.188 Acc@5 95.308
Accuracy of the network on the 50000 test images: 80.2%
 * Acc@1 80.822 Acc@5 95.676
Accuracy of the ema network on the 50000 test images: 80.8%
--
EPOCH 207
 * Acc@1 80.250 Acc@5 95.382
Accuracy of the network on the 50000 test images: 80.3%
 * Acc@1 80.850 Acc@5 95.674
Accuracy of the ema network on the 50000 test images: 80.8%
--
EPOCH 208
 * Acc@1 80.228 Acc@5 95.370
Accuracy of the network on the 50000 test images: 80.2%
 * Acc@1 80.884 Acc@5 95.696
Accuracy of the ema network on the 50000 test images: 80.9%
--
EPOCH 209
 * Acc@1 80.380 Acc@5 95.448
Accuracy of the network on the 50000 test images: 80.4%
 * Acc@1 80.900 Acc@5 95.702
Accuracy of the ema network on the 50000 test images: 80.9%
--
EPOCH 210
 * Acc@1 80.278 Acc@5 95.410
Accuracy of the network on the 50000 test images: 80.3%
 * Acc@1 80.942 Acc@5 95.722
Accuracy of the ema network on the 50000 test images: 80.9%
--
EPOCH 211
 * Acc@1 80.376 Acc@5 95.536
Accuracy of the network on the 50000 test images: 80.4%
 * Acc@1 80.958 Acc@5 95.712
Accuracy of the ema network on the 50000 test images: 81.0%
--
EPOCH 212
 * Acc@1 80.274 Acc@5 95.436
Accuracy of the network on the 50000 test images: 80.3%
 * Acc@1 80.976 Acc@5 95.728
Accuracy of the ema network on the 50000 test images: 81.0%
--
EPOCH 213
 * Acc@1 80.304 Acc@5 95.376
Accuracy of the network on the 50000 test images: 80.3%
 * Acc@1 80.996 Acc@5 95.722
Accuracy of the ema network on the 50000 test images: 81.0%
--
EPOCH 214
 * Acc@1 80.378 Acc@5 95.496
Accuracy of the network on the 50000 test images: 80.4%
 * Acc@1 81.008 Acc@5 95.748
Accuracy of the ema network on the 50000 test images: 81.0%
--
EPOCH 215
 * Acc@1 80.536 Acc@5 95.486
Accuracy of the network on the 50000 test images: 80.5%
 * Acc@1 81.048 Acc@5 95.766
Accuracy of the ema network on the 50000 test images: 81.0%
--
EPOCH 216
 * Acc@1 80.434 Acc@5 95.466
Accuracy of the network on the 50000 test images: 80.4%
 * Acc@1 81.068 Acc@5 95.762
Accuracy of the ema network on the 50000 test images: 81.1%
--
EPOCH 217
 * Acc@1 80.560 Acc@5 95.534
Accuracy of the network on the 50000 test images: 80.6%
 * Acc@1 81.090 Acc@5 95.796
Accuracy of the ema network on the 50000 test images: 81.1%
--
EPOCH 218
 * Acc@1 80.500 Acc@5 95.490
Accuracy of the network on the 50000 test images: 80.5%
 * Acc@1 81.102 Acc@5 95.794
Accuracy of the ema network on the 50000 test images: 81.1%
--
EPOCH 219
 * Acc@1 80.776 Acc@5 95.638
Accuracy of the network on the 50000 test images: 80.8%
 * Acc@1 81.132 Acc@5 95.796
Accuracy of the ema network on the 50000 test images: 81.1%
--
EPOCH 220
 * Acc@1 80.658 Acc@5 95.582
Accuracy of the network on the 50000 test images: 80.7%
 * Acc@1 81.146 Acc@5 95.820
Accuracy of the ema network on the 50000 test images: 81.1%
--
EPOCH 221
 * Acc@1 80.714 Acc@5 95.550
Accuracy of the network on the 50000 test images: 80.7%
 * Acc@1 81.172 Acc@5 95.816
Accuracy of the ema network on the 50000 test images: 81.2%
--
EPOCH 222
 * Acc@1 80.800 Acc@5 95.570
Accuracy of the network on the 50000 test images: 80.8%
 * Acc@1 81.194 Acc@5 95.826
Accuracy of the ema network on the 50000 test images: 81.2%
--
EPOCH 223
 * Acc@1 80.696 Acc@5 95.664
Accuracy of the network on the 50000 test images: 80.7%
 * Acc@1 81.216 Acc@5 95.846
Accuracy of the ema network on the 50000 test images: 81.2%
--
EPOCH 224
 * Acc@1 80.724 Acc@5 95.612
Accuracy of the network on the 50000 test images: 80.7%
 * Acc@1 81.230 Acc@5 95.852
Accuracy of the ema network on the 50000 test images: 81.2%
--
EPOCH 225
 * Acc@1 80.818 Acc@5 95.632
Accuracy of the network on the 50000 test images: 80.8%
 * Acc@1 81.276 Acc@5 95.850
Accuracy of the ema network on the 50000 test images: 81.3%
--
EPOCH 226
 * Acc@1 81.030 Acc@5 95.652
Accuracy of the network on the 50000 test images: 81.0%
 * Acc@1 81.272 Acc@5 95.858
Accuracy of the ema network on the 50000 test images: 81.3%
--
EPOCH 227
 * Acc@1 80.932 Acc@5 95.722
Accuracy of the network on the 50000 test images: 80.9%
 * Acc@1 81.286 Acc@5 95.858
Accuracy of the ema network on the 50000 test images: 81.3%
--
EPOCH 228
 * Acc@1 80.912 Acc@5 95.750
Accuracy of the network on the 50000 test images: 80.9%
 * Acc@1 81.296 Acc@5 95.862
Accuracy of the ema network on the 50000 test images: 81.3%
--
EPOCH 229
 * Acc@1 80.954 Acc@5 95.652
Accuracy of the network on the 50000 test images: 81.0%
 * Acc@1 81.320 Acc@5 95.840
Accuracy of the ema network on the 50000 test images: 81.3%
--
EPOCH 230
 * Acc@1 81.166 Acc@5 95.708
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.380 Acc@5 95.846
Accuracy of the ema network on the 50000 test images: 81.4%
--
EPOCH 231
 * Acc@1 81.092 Acc@5 95.776
Accuracy of the network on the 50000 test images: 81.1%
 * Acc@1 81.398 Acc@5 95.844
Accuracy of the ema network on the 50000 test images: 81.4%
--
EPOCH 232
 * Acc@1 81.044 Acc@5 95.780
Accuracy of the network on the 50000 test images: 81.0%
 * Acc@1 81.406 Acc@5 95.856
Accuracy of the ema network on the 50000 test images: 81.4%
--
EPOCH 233
 * Acc@1 80.966 Acc@5 95.710
Accuracy of the network on the 50000 test images: 81.0%
 * Acc@1 81.446 Acc@5 95.866
Accuracy of the ema network on the 50000 test images: 81.4%
--
EPOCH 234
 * Acc@1 81.142 Acc@5 95.726
Accuracy of the network on the 50000 test images: 81.1%
 * Acc@1 81.506 Acc@5 95.890
Accuracy of the ema network on the 50000 test images: 81.5%
--
EPOCH 235
 * Acc@1 81.206 Acc@5 95.716
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.498 Acc@5 95.908
Accuracy of the ema network on the 50000 test images: 81.5%
--
EPOCH 236
 * Acc@1 81.250 Acc@5 95.746
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.500 Acc@5 95.890
Accuracy of the ema network on the 50000 test images: 81.5%
--
EPOCH 237
 * Acc@1 81.152 Acc@5 95.774
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.546 Acc@5 95.912
Accuracy of the ema network on the 50000 test images: 81.5%
--
EPOCH 238
 * Acc@1 81.246 Acc@5 95.814
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.570 Acc@5 95.916
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 239
 * Acc@1 81.248 Acc@5 95.832
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.574 Acc@5 95.910
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 240
 * Acc@1 81.094 Acc@5 95.786
Accuracy of the network on the 50000 test images: 81.1%
 * Acc@1 81.584 Acc@5 95.924
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 241
 * Acc@1 81.404 Acc@5 95.792
Accuracy of the network on the 50000 test images: 81.4%
 * Acc@1 81.620 Acc@5 95.930
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 242
 * Acc@1 81.458 Acc@5 95.808
Accuracy of the network on the 50000 test images: 81.5%
 * Acc@1 81.630 Acc@5 95.942
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 243
 * Acc@1 81.276 Acc@5 95.854
Accuracy of the network on the 50000 test images: 81.3%
 * Acc@1 81.628 Acc@5 95.946
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 244
 * Acc@1 81.444 Acc@5 95.868
Accuracy of the network on the 50000 test images: 81.4%
 * Acc@1 81.626 Acc@5 95.954
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 245
 * Acc@1 81.384 Acc@5 95.856
Accuracy of the network on the 50000 test images: 81.4%
 * Acc@1 81.640 Acc@5 95.960
Accuracy of the ema network on the 50000 test images: 81.6%
--
EPOCH 246
 * Acc@1 81.428 Acc@5 95.814
Accuracy of the network on the 50000 test images: 81.4%
 * Acc@1 81.664 Acc@5 95.982
Accuracy of the ema network on the 50000 test images: 81.7%
--
EPOCH 247
 * Acc@1 81.194 Acc@5 95.764
Accuracy of the network on the 50000 test images: 81.2%
 * Acc@1 81.712 Acc@5 96.000
Accuracy of the ema network on the 50000 test images: 81.7%
--
EPOCH 248
 * Acc@1 81.476 Acc@5 95.828
Accuracy of the network on the 50000 test images: 81.5%
 * Acc@1 81.732 Acc@5 96.000
Accuracy of the ema network on the 50000 test images: 81.7%
--
EPOCH 249
 * Acc@1 81.472 Acc@5 95.866
Accuracy of the network on the 50000 test images: 81.5%
 * Acc@1 81.750 Acc@5 96.008
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 250
 * Acc@1 81.606 Acc@5 95.898
Accuracy of the network on the 50000 test images: 81.6%
 * Acc@1 81.726 Acc@5 96.008
Accuracy of the ema network on the 50000 test images: 81.7%
--
EPOCH 251
 * Acc@1 81.626 Acc@5 95.902
Accuracy of the network on the 50000 test images: 81.6%
 * Acc@1 81.760 Acc@5 96.002
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 252
 * Acc@1 81.664 Acc@5 95.926
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.778 Acc@5 96.002
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 253
 * Acc@1 81.514 Acc@5 95.828
Accuracy of the network on the 50000 test images: 81.5%
 * Acc@1 81.800 Acc@5 95.994
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 254
 * Acc@1 81.562 Acc@5 95.878
Accuracy of the network on the 50000 test images: 81.6%
 * Acc@1 81.816 Acc@5 95.990
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 255
 * Acc@1 81.660 Acc@5 95.956
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.842 Acc@5 95.998
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 256
 * Acc@1 81.626 Acc@5 95.872
Accuracy of the network on the 50000 test images: 81.6%
 * Acc@1 81.852 Acc@5 96.004
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 257
 * Acc@1 81.748 Acc@5 95.868
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.868 Acc@5 96.006
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 258
 * Acc@1 81.728 Acc@5 95.946
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.860 Acc@5 96.026
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 259
 * Acc@1 81.746 Acc@5 95.964
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.858 Acc@5 96.022
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 260
 * Acc@1 81.568 Acc@5 95.868
Accuracy of the network on the 50000 test images: 81.6%
 * Acc@1 81.862 Acc@5 96.016
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 261
 * Acc@1 81.692 Acc@5 95.998
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.848 Acc@5 96.038
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 262
 * Acc@1 81.738 Acc@5 96.032
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.862 Acc@5 96.024
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 263
 * Acc@1 81.750 Acc@5 96.030
Accuracy of the network on the 50000 test images: 81.8%
 * Acc@1 81.860 Acc@5 96.052
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 264
 * Acc@1 81.780 Acc@5 96.020
Accuracy of the network on the 50000 test images: 81.8%
 * Acc@1 81.872 Acc@5 96.066
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 265
 * Acc@1 81.730 Acc@5 96.024
Accuracy of the network on the 50000 test images: 81.7%
 * Acc@1 81.868 Acc@5 96.072
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 266
 * Acc@1 81.790 Acc@5 95.998
Accuracy of the network on the 50000 test images: 81.8%
 * Acc@1 81.846 Acc@5 96.076
Accuracy of the ema network on the 50000 test images: 81.8%
--
EPOCH 267
 * Acc@1 81.770 Acc@5 96.034
Accuracy of the network on the 50000 test images: 81.8%
 * Acc@1 81.886 Acc@5 96.092
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 268
 * Acc@1 81.882 Acc@5 96.042
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 81.900 Acc@5 96.100
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 269
 * Acc@1 81.920 Acc@5 96.036
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 81.888 Acc@5 96.094
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 270
 * Acc@1 81.896 Acc@5 95.978
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 81.906 Acc@5 96.082
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 271
 * Acc@1 81.976 Acc@5 96.000
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 81.920 Acc@5 96.070
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 272
 * Acc@1 81.968 Acc@5 96.016
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 81.930 Acc@5 96.078
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 273
 * Acc@1 81.852 Acc@5 95.962
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 81.910 Acc@5 96.082
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 274
 * Acc@1 81.914 Acc@5 96.010
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 81.936 Acc@5 96.086
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 275
 * Acc@1 81.958 Acc@5 96.036
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 81.946 Acc@5 96.082
Accuracy of the ema network on the 50000 test images: 81.9%
--
EPOCH 276
 * Acc@1 82.020 Acc@5 96.028
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 81.956 Acc@5 96.086
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 277
 * Acc@1 81.918 Acc@5 96.052
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 81.968 Acc@5 96.084
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 278
 * Acc@1 81.986 Acc@5 96.076
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 81.996 Acc@5 96.084
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 279
 * Acc@1 82.032 Acc@5 96.104
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 82.010 Acc@5 96.094
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 280
 * Acc@1 81.938 Acc@5 96.066
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 82.022 Acc@5 96.102
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 281
 * Acc@1 82.046 Acc@5 96.124
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 82.034 Acc@5 96.104
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 282
 * Acc@1 81.946 Acc@5 96.034
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 82.050 Acc@5 96.106
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 283
 * Acc@1 81.990 Acc@5 96.118
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 82.046 Acc@5 96.106
Accuracy of the ema network on the 50000 test images: 82.0%
--
EPOCH 284
 * Acc@1 81.918 Acc@5 96.052
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 82.052 Acc@5 96.086
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 285
 * Acc@1 82.098 Acc@5 96.092
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.070 Acc@5 96.100
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 286
 * Acc@1 82.142 Acc@5 96.112
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.070 Acc@5 96.108
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 287
 * Acc@1 82.106 Acc@5 96.120
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.074 Acc@5 96.112
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 288
 * Acc@1 81.918 Acc@5 96.096
Accuracy of the network on the 50000 test images: 81.9%
 * Acc@1 82.096 Acc@5 96.116
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 289
 * Acc@1 82.086 Acc@5 96.124
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.092 Acc@5 96.120
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 290
 * Acc@1 82.052 Acc@5 96.092
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.096 Acc@5 96.124
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 291
 * Acc@1 82.090 Acc@5 96.112
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.094 Acc@5 96.120
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 292
 * Acc@1 82.126 Acc@5 96.070
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.140 Acc@5 96.122
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 293
 * Acc@1 82.064 Acc@5 96.104
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.122 Acc@5 96.122
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 294
 * Acc@1 82.100 Acc@5 96.104
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.130 Acc@5 96.116
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 295
 * Acc@1 82.060 Acc@5 96.072
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.118 Acc@5 96.114
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 296
 * Acc@1 82.072 Acc@5 96.142
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.128 Acc@5 96.126
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 297
 * Acc@1 82.012 Acc@5 96.070
Accuracy of the network on the 50000 test images: 82.0%
 * Acc@1 82.122 Acc@5 96.128
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 298
 * Acc@1 82.156 Acc@5 96.138
Accuracy of the network on the 50000 test images: 82.2%
 * Acc@1 82.110 Acc@5 96.122
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 299
 * Acc@1 82.150 Acc@5 96.136
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.124 Acc@5 96.130
Accuracy of the ema network on the 50000 test images: 82.1%
--
EPOCH 300
 * Acc@1 82.116 Acc@5 96.140
Accuracy of the network on the 50000 test images: 82.1%
 * Acc@1 82.136 Acc@5 96.136
Accuracy of the ema network on the 50000 test images: 82.1%