Cannot reproduce results on ImageNet1K

[techmn]

Hi,
Thanks for sharing the good work.
I was reproducing the results on ImageNet1k using the hyperparameters settings provided on the repo but I am not able to reproduce the results of Acc1: 67.9 for SeaFormer_T. Instead I get the Acc1 value as 66.23.
Here are the parameter values from the log file

aa: rand-m9-mstd0.5
amp: true
apex_amp: false
aug_repeats: 0
aug_splits: 0
batch_size: 128
bce_loss: false
bce_target_thresh: null
bn_eps: null
bn_momentum: null
channels_last: false
checkpoint_hist: 10
class_map: ''
clip_grad: null
clip_mode: norm
color_jitter: 0.4
cooldown_epochs: 10
crop_pct: null
cutmix: 0.0
cutmix_minmax: null
data_dir: ./imagenet_1k
dataset: ''
dataset_download: false
decay_epochs: 2.4
decay_rate: 0.973
dist_bn: reduce
drop: 0.2
drop_block: null
drop_connect: 0.2
drop_path: null
epoch_repeats: 0.0
epochs: 600
eval_metric: top1
experiment: SeaFormer_T
fuser: ''
gp: null
grad_checkpointing: false
hflip: 0.5
img_size: 224
initial_checkpoint: ''
input_size:

• 3
• 224
• 224

interpolation: ''
jsd_loss: false
layer_decay: null
local_rank: 0
log_interval: 50
log_wandb: false
lr: 0.064
lr_cycle_decay: 0.5
lr_cycle_limit: 1
lr_cycle_mul: 1.0
lr_k_decay: 1.0
lr_noise:

• 0.42

lr_noise_pct: 0.67
lr_noise_std: 1.0
mean: null
min_lr: 1.0e-06
mixup: 0.0
mixup_mode: batch
mixup_off_epoch: 0
mixup_prob: 1.0
mixup_switch_prob: 0.5
model: SeaFormer_T
model_ema: true
model_ema_decay: 0.9999
model_ema_force_cpu: false
momentum: 0.9
native_amp: false
no_aug: false
no_ddp_bb: false
no_prefetcher: false
no_resume_opt: false
num_classes: 1000
opt: adamw
opt_betas: null
opt_eps: 0.001
output: ./output_dir
patience_epochs: 10
pin_mem: false
pretrained: false
ratio:

• 0.75
• 1.3333333333333333

recount: 1
recovery_interval: 0
remode: pixel
reprob: 0.2
resplit: false
resume: ''
save_images: false
scale:

• 0.08
• 1.0

sched: cosine
seed: 42
smoothing: 0.1
split_bn: false
start_epoch: null
std: null
sync_bn: false
torchscript: false
train_interpolation: random
train_split: train
tta: 0
use_multi_epochs_loader: false
val_split: validation
validation_batch_size: null
vflip: 0.0
warmup_epochs: 10
warmup_lr: 1.0e-06
weight_decay: 2.0e-05
worker_seeding: all
workers: 7

Could you please tell me if the parameters values are correct and what am I doing wrong here. As recommended in repo I utilized 8 GPUs but I am not able to reproduce the results. Same is the case with SeaFormer_S.