_tiny_test_README.md

Hparams for tiny test timm models, all pretty similar, vit3 extended the schedule from the default 1600 epochs to 1800. All based on MobileNetv4 template w/ adamw reduced beta1, grinding out the training. Some of the smallest models had a bit less AA magnitude (3), slightly higher capacity ones increased to m5 or m6.

test_convnext3_e1600.yaml

aa: rand-m3-inc1-mstd101
amp: true
amp_dtype: float16
amp_impl: native
aug_repeats: 0
aug_splits: 0
batch_size: 1024
bce_loss: false
bce_pos_weight: null
bce_sum: false
bce_target_thresh: null
bn_eps: null
bn_momentum: null
channels_last: true
checkpoint_hist: 10
class_map: ''
clip_grad: 5.0
clip_mode: norm
color_jitter: 0.4
color_jitter_prob: null
cooldown_epochs: 0
crop_pct: null
cutmix: 0.0
cutmix_minmax: null
data: /data/imagenet/
data_dir: null
dataset: ''
dataset_download: false
decay_epochs: 90
decay_milestones:
- 90
- 180
- 270
decay_rate: 0.1
device: cuda
device_modules: null
dist_bn: reduce
drop: 0.3
drop_block: null
drop_connect: null
drop_path: 0.05
epoch_repeats: 0.0
epochs: 1600
eval_metric: top1
experiment: ''
fast_norm: false
fuser: ''
gaussian_blur_prob: null
gp: null
grad_accum_steps: 1
grad_checkpointing: false
grayscale_prob: 0.05
head_init_bias: null
head_init_scale: 0.0
hflip: 0.5
img_size: null
in_chans: null
initial_checkpoint: ''
input_img_mode: null
input_key: null
input_size: null
interpolation: ''
jsd_loss: false
layer_decay: null
local_rank: 0
log_interval: 50
log_wandb: false
lr: 0.00167
lr_base: 0.1
lr_base_scale: ''
lr_base_size: 256
lr_cycle_decay: 0.5
lr_cycle_limit: 1
lr_cycle_mul: 1.0
lr_k_decay: 1.0
lr_noise: null
lr_noise_pct: 0.67
lr_noise_std: 1.0
mean: null
min_lr: 0.0
mixup: 0.0
mixup_mode: batch
mixup_off_epoch: 0
mixup_prob: 1.0
mixup_switch_prob: 0.5
model: test_convnext3
model_ema: true
model_ema_decay: 0.9999
model_ema_force_cpu: false
model_ema_warmup: true
model_kwargs: {}
momentum: 0.9
no_aug: false
no_ddp_bb: false
no_prefetcher: false
no_resume_opt: false
num_classes: 1000
opt: adamw
opt_betas: null
opt_eps: null
opt_kwargs: {}
output: ''
patience_epochs: 10
pin_mem: false
pretrained: false
pretrained_path: null
ratio:
- 0.75
- 1.3333333333333333
recount: 1
recovery_interval: 0
remode: pixel
reprob: 0.2
resplit: false
resume: ''
save_images: false
scale:
- 0.8
- 1.0
sched: cosine
sched_on_updates: true
seed: 42
smoothing: 0.1
split_bn: false
start_epoch: null
std: null
sync_bn: false
synchronize_step: false
target_key: null
torchcompile: inductor
torchscript: false
train_crop_mode: rkrr
train_interpolation: random
train_num_samples: null
train_split: train
tta: 0
use_multi_epochs_loader: false
val_num_samples: null
val_split: validation
validation_batch_size: null
vflip: 0.0
warmup_epochs: 20
warmup_lr: 0.0
warmup_prefix: false
weight_decay: 0.05
worker_seeding: all
workers: 8

test_vit2_e1600.yaml

aa: rand-m5-inc1-mstd101
amp: true
amp_dtype: float16
amp_impl: native
aug_repeats: 0
aug_splits: 0
batch_size: 1024
bce_loss: false
bce_pos_weight: null
bce_sum: false
bce_target_thresh: null
bn_eps: null
bn_momentum: null
channels_last: true
checkpoint_hist: 10
class_map: ''
clip_grad: 5.0
clip_mode: norm
color_jitter: 0.4
color_jitter_prob: null
cooldown_epochs: 0
crop_pct: null
cutmix: 0.0
cutmix_minmax: null
data: /data/imagenet/
data_dir: null
dataset: ''
dataset_download: false
decay_epochs: 90
decay_milestones:
- 90
- 180
- 270
decay_rate: 0.1
device: cuda
device_modules: null
dist_bn: reduce
drop: 0.3
drop_block: null
drop_connect: null
drop_path: 0.05
epoch_repeats: 0.0
epochs: 1600
eval_metric: top1
experiment: ''
fast_norm: false
fuser: ''
gaussian_blur_prob: null
gp: null
grad_accum_steps: 1
grad_checkpointing: false
grayscale_prob: 0.05
head_init_bias: null
head_init_scale: 0.0
hflip: 0.5
img_size: null
in_chans: null
initial_checkpoint: ''
input_img_mode: null
input_key: null
input_size: null
interpolation: ''
jsd_loss: false
layer_decay: null
local_rank: 0
log_interval: 50
log_wandb: false
lr: 0.00167
lr_base: 0.1
lr_base_scale: ''
lr_base_size: 256
lr_cycle_decay: 0.5
lr_cycle_limit: 1
lr_cycle_mul: 1.0
lr_k_decay: 1.0
lr_noise: null
lr_noise_pct: 0.67
lr_noise_std: 1.0
mean: null
min_lr: 0.0
mixup: 0.0
mixup_mode: batch
mixup_off_epoch: 0
mixup_prob: 1.0
mixup_switch_prob: 0.5
model: test_vit2
model_ema: true
model_ema_decay: 0.9999
model_ema_force_cpu: false
model_ema_warmup: true
model_kwargs: {}
momentum: 0.9
no_aug: false
no_ddp_bb: false
no_prefetcher: false
no_resume_opt: false
num_classes: 1000
opt: adamw
opt_betas: null
opt_eps: null
opt_kwargs: {}
output: ''
patience_epochs: 10
pin_mem: false
pretrained: false
pretrained_path: null
ratio:
- 0.75
- 1.3333333333333333
recount: 1
recovery_interval: 0
remode: pixel
reprob: 0.2
resplit: false
resume: ''
save_images: false
scale:
- 0.8
- 1.0
sched: cosine
sched_on_updates: true
seed: 42
smoothing: 0.1
split_bn: false
start_epoch: null
std: null
sync_bn: false
synchronize_step: false
target_key: null
torchcompile: inductor
torchscript: false
train_crop_mode: rkrr
train_interpolation: random
train_num_samples: null
train_split: train
tta: 0
use_multi_epochs_loader: false
val_num_samples: null
val_split: validation
validation_batch_size: null
vflip: 0.0
warmup_epochs: 20
warmup_lr: 0.0
warmup_prefix: false
weight_decay: 0.05
worker_seeding: all
workers: 8

test_vit3_e1800.yaml

aa: rand-m6-inc1-mstd101
amp: true
amp_dtype: float16
amp_impl: native
aug_repeats: 0
aug_splits: 0
batch_size: 1024
bce_loss: false
bce_pos_weight: null
bce_sum: false
bce_target_thresh: null
bn_eps: null
bn_momentum: null
channels_last: true
checkpoint_hist: 10
class_map: ''
clip_grad: 5.0
clip_mode: norm
color_jitter: 0.4
color_jitter_prob: null
cooldown_epochs: 0
crop_pct: 0.9
cutmix: 0.0
cutmix_minmax: null
data: /data/imagenet/
data_dir: null
dataset: ''
dataset_download: false
decay_epochs: 90
decay_milestones:
- 90
- 180
- 270
decay_rate: 0.1
device: cuda
device_modules: null
dist_bn: reduce
drop: 0.3
drop_block: null
drop_connect: null
drop_path: 0.05
epoch_repeats: 0.0
epochs: 1800
eval_metric: top1
experiment: ''
fast_norm: false
fuser: ''
gaussian_blur_prob: null
gp: null
grad_accum_steps: 1
grad_checkpointing: false
grayscale_prob: 0.05
head_init_bias: null
head_init_scale: 0.0
hflip: 0.5
img_size: null
in_chans: null
initial_checkpoint: ''
input_img_mode: null
input_key: null
input_size: null
interpolation: ''
jsd_loss: false
layer_decay: null
local_rank: 0
log_interval: 50
log_wandb: false
lr: 0.00167
lr_base: 0.1
lr_base_scale: ''
lr_base_size: 256
lr_cycle_decay: 0.5
lr_cycle_limit: 1
lr_cycle_mul: 1.0
lr_k_decay: 1.0
lr_noise: null
lr_noise_pct: 0.67
lr_noise_std: 1.0
mean: null
min_lr: 0.0
mixup: 0.0
mixup_mode: batch
mixup_off_epoch: 0
mixup_prob: 1.0
mixup_switch_prob: 0.5
model: test_vit3
model_ema: true
model_ema_decay: 0.9999
model_ema_force_cpu: false
model_ema_warmup: true
model_kwargs: {}
momentum: 0.9
no_aug: false
no_ddp_bb: false
no_prefetcher: false
no_resume_opt: false
num_classes: 1000
opt: adamw
opt_betas: null
opt_eps: null
opt_kwargs: {}
output: ''
patience_epochs: 10
pin_mem: false
pretrained: false
pretrained_path: null
ratio:
- 0.75
- 1.3333333333333333
recount: 1
recovery_interval: 0
remode: pixel
reprob: 0.2
resplit: false
resume: ''
save_images: false
scale:
- 0.8
- 1.0
sched: cosine
sched_on_updates: true
seed: 42
smoothing: 0.1
split_bn: false
start_epoch: null
std: null
sync_bn: false
synchronize_step: false
target_key: null
torchcompile: inductor
torchscript: false
train_crop_mode: rkrr
train_interpolation: random
train_num_samples: null
train_split: train
tta: 0
use_multi_epochs_loader: false
val_num_samples: null
val_split: validation
validation_batch_size: null
vflip: 0.0
warmup_epochs: 20
warmup_lr: 0.0
warmup_prefix: false
weight_decay: 0.05
worker_seeding: all
workers: 8