Update auto_scale_lr fields

configs/_base_/schedules/cifar10_bs128.py

@@ -9,3 +9,7 @@ param_scheduler = dict(
 train_cfg = dict(by_epoch=True, max_epochs=200, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=128)

configs/_base_/schedules/cub_bs64.py

@@ -28,3 +28,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=64)

configs/_base_/schedules/imagenet_bs1024_adamw_conformer.py

@@ -37,3 +37,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=300, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=1024)

configs/_base_/schedules/imagenet_bs1024_adamw_swin.py

@@ -41,3 +41,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=300, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=1024)

configs/_base_/schedules/imagenet_bs1024_coslr.py

@@ -12,3 +12,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=1024)

configs/_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py

@@ -20,3 +20,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=300, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=1024)

configs/_base_/schedules/imagenet_bs2048.py

@@ -15,3 +15,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=2048)

configs/_base_/schedules/imagenet_bs2048_AdamW.py

@@ -36,3 +36,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=300, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=2048)

configs/_base_/schedules/imagenet_bs2048_coslr.py

@@ -29,3 +29,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=2048)

configs/_base_/schedules/imagenet_bs2048_rsb.py

@@ -26,3 +26,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=2048)

configs/_base_/schedules/imagenet_bs256.py

@@ -10,3 +10,7 @@ param_scheduler = dict(
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=256)

configs/_base_/schedules/imagenet_bs256_140e.py

@@ -10,3 +10,7 @@ param_scheduler = dict(
 train_cfg = dict(by_epoch=True, max_epochs=140, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=256)

configs/_base_/schedules/imagenet_bs256_200e_coslr_warmup.py

@@ -28,3 +28,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=200, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=256)

configs/_base_/schedules/imagenet_bs256_coslr.py

@@ -10,3 +10,7 @@ param_scheduler = dict(
 train_cfg = dict(by_epoch=True, max_epochs=100, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=256)

configs/_base_/schedules/imagenet_bs256_epochstep.py

@@ -9,3 +9,7 @@ param_scheduler = dict(type='StepLR', by_epoch=True, step_size=1, gamma=0.98)
 train_cfg = dict(by_epoch=True, max_epochs=300, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=256)

configs/_base_/schedules/imagenet_bs4096_AdamW.py

@@ -33,3 +33,7 @@ param_scheduler = [
 train_cfg = dict(by_epoch=True, max_epochs=300, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# based on the actual training batch size.
+auto_scale_lr = dict(base_batch_size=4096)

configs/conformer/conformer-base-p16_8xb128_in1k.py

@@ -6,8 +6,3 @@ _base_ = [
 ]
 
 train_dataloader = dict(batch_size=128)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/conformer/conformer-small-p16_8xb128_in1k.py

@@ -6,8 +6,3 @@ _base_ = [
 ]
 
 train_dataloader = dict(batch_size=128)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/conformer/conformer-small-p32_8xb128_in1k.py

@@ -6,8 +6,3 @@ _base_ = [
 ]
 
 train_dataloader = dict(batch_size=128)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/conformer/conformer-tiny-p16_8xb128_in1k.py

@@ -6,8 +6,3 @@ _base_ = [
 ]
 
 train_dataloader = dict(batch_size=128)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/convmixer/convmixer-1024-20_10xb64_in1k.py

@@ -33,7 +33,7 @@ param_scheduler = [
 
 train_cfg = dict(by_epoch=True, max_epochs=150)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (10 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=640)

configs/convmixer/convmixer-1536-20_10xb64_in1k.py

@@ -33,7 +33,7 @@ param_scheduler = [
 
 train_cfg = dict(by_epoch=True, max_epochs=150)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (10 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=640)

configs/convmixer/convmixer-768-32_10xb64_in1k.py

@@ -13,7 +13,7 @@ optim_wrapper = dict(
 
 train_cfg = dict(by_epoch=True, max_epochs=300)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (10 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=640)

configs/convnext/convnext-base_32xb128_in1k.py

@@ -17,7 +17,7 @@ optim_wrapper = dict(
 # runtime setting
 custom_hooks = [dict(type='EMAHook', momentum=4e-5, priority='ABOVE_NORMAL')]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/convnext/convnext-large_64xb64_in1k.py

@@ -17,7 +17,7 @@ optim_wrapper = dict(
 # runtime setting
 custom_hooks = [dict(type='EMAHook', momentum=4e-5, priority='ABOVE_NORMAL')]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (64 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/convnext/convnext-small_32xb128_in1k.py

@@ -17,7 +17,7 @@ optim_wrapper = dict(
 # runtime setting
 custom_hooks = [dict(type='EMAHook', momentum=4e-5, priority='ABOVE_NORMAL')]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/convnext/convnext-tiny_32xb128_in1k.py

@@ -17,7 +17,7 @@ optim_wrapper = dict(
 # runtime setting
 custom_hooks = [dict(type='EMAHook', momentum=4e-5, priority='ABOVE_NORMAL')]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/convnext/convnext-xlarge_64xb64_in1k.py

@@ -17,7 +17,7 @@ optim_wrapper = dict(
 # runtime setting
 custom_hooks = [dict(type='EMAHook', momentum=4e-5, priority='ABOVE_NORMAL')]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (64 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/cspnet/cspdarknet50_8xb32_in1k.py

@@ -43,8 +43,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/cspnet/cspresnet50_8xb32_in1k.py

@@ -43,8 +43,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/cspnet/cspresnext50_8xb32_in1k.py

@@ -43,8 +43,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/deit/deit-base-distilled_ft-16xb32_in1k-384px.py

@@ -7,8 +7,3 @@ model = dict(
     # Change to the path of the pretrained model
     # init_cfg=dict(type='Pretrained', checkpoint=''),
 )
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (16 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=512)

configs/deit/deit-base-distilled_pt-16xb64_in1k.py

@@ -8,8 +8,3 @@ model = dict(
 
 # dataset settings
 train_dataloader = dict(batch_size=64)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (16 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/deit/deit-base_ft-16xb32_in1k-384px.py

@@ -31,7 +31,7 @@ train_dataloader = dict(batch_size=32)
 # schedule settings
 optim_wrapper = dict(clip_grad=dict(max_norm=1.0))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (16 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/deit/deit-base_pt-16xb64_in1k.py

@@ -12,8 +12,3 @@ train_dataloader = dict(batch_size=64)
 
 # runtime settings
 custom_hooks = [dict(type='EMAHook', momentum=4e-5, priority='ABOVE_NORMAL')]
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (16 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/deit/deit-small-distilled_pt-4xb256_in1k.py

@@ -5,8 +5,3 @@ model = dict(
     backbone=dict(type='DistilledVisionTransformer', arch='deit-small'),
     head=dict(type='DeiTClsHead', in_channels=384),
 )
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (256 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/deit/deit-small_pt-4xb256_in1k.py

@@ -46,8 +46,3 @@ optim_wrapper = dict(
         }),
     clip_grad=dict(max_norm=5.0),
 )
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (256 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/deit/deit-tiny-distilled_pt-4xb256_in1k.py

@@ -5,8 +5,3 @@ model = dict(
     backbone=dict(type='DistilledVisionTransformer', arch='deit-tiny'),
     head=dict(type='DeiTClsHead', in_channels=192),
 )
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (256 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/deit/deit-tiny_pt-4xb256_in1k.py

@@ -5,8 +5,3 @@ model = dict(
     backbone=dict(type='VisionTransformer', arch='deit-tiny'),
     head=dict(type='VisionTransformerClsHead', in_channels=192),
 )
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (256 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/densenet/densenet121_4xb256_in1k.py

@@ -11,7 +11,7 @@ train_dataloader = dict(batch_size=256)
 # schedule settings
 train_cfg = dict(by_epoch=True, max_epochs=90)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (256 samples per GPU)
 auto_scale_lr = dict(base_batch_size=1024)

configs/densenet/densenet161_4xb256_in1k.py

@@ -11,7 +11,7 @@ train_dataloader = dict(batch_size=256)
 # schedule settings
 train_cfg = dict(by_epoch=True, max_epochs=90)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (256 samples per GPU)
 auto_scale_lr = dict(base_batch_size=1024)

configs/densenet/densenet169_4xb256_in1k.py

@@ -11,7 +11,7 @@ train_dataloader = dict(batch_size=256)
 # schedule settings
 train_cfg = dict(by_epoch=True, max_epochs=90)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (256 samples per GPU)
 auto_scale_lr = dict(base_batch_size=1024)

configs/densenet/densenet201_4xb256_in1k.py

@@ -11,7 +11,7 @@ train_dataloader = dict(batch_size=256)
 # schedule settings
 train_cfg = dict(by_epoch=True, max_epochs=90)
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (256 samples per GPU)
 auto_scale_lr = dict(base_batch_size=1024)

configs/efficientnet/efficientnet-b0_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b0_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b1_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b1_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b2_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b2_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b3_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b3_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b4_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b4_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b5_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b5_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b6_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b6_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b7_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b7_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b8_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-b8_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-em_8xb32-01norm_in1k.py

@@ -29,8 +29,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/efficientnet/efficientnet-es_8xb32-01norm_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/hrnet/hrnet-w18_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/hrnet/hrnet-w30_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/hrnet/hrnet-w32_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/hrnet/hrnet-w40_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/hrnet/hrnet-w44_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/hrnet/hrnet-w48_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/hrnet/hrnet-w64_4xb32_in1k.py

@@ -5,7 +5,7 @@ _base_ = [
     '../_base_/default_runtime.py'
 ]
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (4 GPUs) x (32 samples per GPU)
 auto_scale_lr = dict(base_batch_size=128)

configs/inception_v3/inception-v3_8xb32_in1k.py

@@ -22,8 +22,3 @@ test_pipeline = [
 train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
 val_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/lenet/lenet5_mnist.py

@@ -85,7 +85,7 @@ load_from = None
 # whether to resume the training of the checkpoint
 resume_from = None
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (1 GPUs) x (128 samples per GPU)
+auto_scale_lr = dict(base_batch_size=128)

configs/mlp_mixer/mlp-mixer-base-p16_64xb64_in1k.py

@@ -6,8 +6,3 @@ _base_ = [
 ]
 
 optim_wrapper = dict(clip_grad=dict(max_norm=1.0))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (64 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=4096)

configs/mlp_mixer/mlp-mixer-large-p16_64xb64_in1k.py

@@ -6,8 +6,3 @@ _base_ = [
 ]
 
 optim_wrapper = dict(clip_grad=dict(max_norm=1.0))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (64 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=4096)

configs/mobilenet_v2/mobilenet-v2_8xb32_in1k.py

@@ -4,8 +4,3 @@ _base_ = [
     '../_base_/schedules/imagenet_bs256_epochstep.py',
     '../_base_/default_runtime.py'
 ]
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/mobilenet_v3/README.md

@@ -18,8 +18,8 @@ We present the next generation of MobileNets based on a combination of complemen
 
 |        Model        | Params(M) | Flops(G) | Top-1 (%) | Top-5 (%) |                                  Config                                  |                                  Download                                  |
 | :-----------------: | :-------: | :------: | :-------: | :-------: | :----------------------------------------------------------------------: | :------------------------------------------------------------------------: |
-| MobileNetV3-Small\* |   2.54    |   0.06   |   67.66   |   87.41   | [config](https://github.com/open-mmlab/mmclassification/blob/master/configs/mobilenet_v3/mobilenet-v3-small_8xb32_in1k.py) | [model](https://download.openmmlab.com/mmclassification/v0/mobilenet_v3/convert/mobilenet_v3_small-8427ecf0.pth) |
-| MobileNetV3-Large\* |   5.48    |   0.23   |   74.04   |   91.34   | [config](https://github.com/open-mmlab/mmclassification/blob/master/configs/mobilenet_v3/mobilenet-v3-large_8xb32_in1k.py) | [model](https://download.openmmlab.com/mmclassification/v0/mobilenet_v3/convert/mobilenet_v3_large-3ea3c186.pth) |
+| MobileNetV3-Small\* |   2.54    |   0.06   |   67.66   |   87.41   | [config](https://github.com/open-mmlab/mmclassification/blob/master/configs/mobilenet_v3/mobilenet-v3-small_8xb128_in1k.py) | [model](https://download.openmmlab.com/mmclassification/v0/mobilenet_v3/convert/mobilenet_v3_small-8427ecf0.pth) |
+| MobileNetV3-Large\* |   5.48    |   0.23   |   74.04   |   91.34   | [config](https://github.com/open-mmlab/mmclassification/blob/master/configs/mobilenet_v3/mobilenet-v3-large_8xb128_in1k.py) | [model](https://download.openmmlab.com/mmclassification/v0/mobilenet_v3/convert/mobilenet_v3_large-3ea3c186.pth) |
 
 *Models with * are converted from [torchvision](https://pytorch.org/vision/stable/_modules/torchvision/models/mobilenetv3.html). The config files of these models are only for validation. We don't ensure these config files' training accuracy and welcome you to contribute your reproduction results.*
 

configs/mobilenet_v3/metafile.yml

@@ -31,7 +31,7 @@ Models:
           Top 5 Accuracy: 87.41
         Task: Image Classification
     Weights: https://download.openmmlab.com/mmclassification/v0/mobilenet_v3/convert/mobilenet_v3_small-8427ecf0.pth
-    Config: configs/mobilenet_v3/mobilenet-v3-small_8xb32_in1k.py
+    Config: configs/mobilenet_v3/mobilenet-v3-small_8xb128_in1k.py
   - Name: mobilenet_v3_large_imagenet
     Metadata:
       FLOPs: 230000000
@@ -44,4 +44,4 @@ Models:
           Top 5 Accuracy: 91.34
         Task: Image Classification
     Weights: https://download.openmmlab.com/mmclassification/v0/mobilenet_v3/convert/mobilenet_v3_large-3ea3c186.pth
-    Config: configs/mobilenet_v3/mobilenet-v3-large_8xb32_in1k.py
+    Config: configs/mobilenet_v3/mobilenet-v3-large_8xb128_in1k.py

configs/mobilenet_v3/mobilenet-v3-large_8xb128_in1k.py

@@ -22,7 +22,7 @@ train_cfg = dict(by_epoch=True, max_epochs=600, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (8 GPUs) x (128 samples per GPU)
+auto_scale_lr = dict(base_batch_size=1024)

configs/mobilenet_v3/mobilenet-v3-small_8xb128_in1k.py

@@ -22,7 +22,7 @@ train_cfg = dict(by_epoch=True, max_epochs=600, val_interval=1)
 val_cfg = dict()
 test_cfg = dict()
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (32 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (8 GPUs) x (128 samples per GPU)
+auto_scale_lr = dict(base_batch_size=1024)

configs/mobilenet_v3/mobilenet-v3-small_8xb16_cifar10.py

@@ -1,5 +1,5 @@
 _base_ = [
-    '../_base_/models/mobilenet-v3-small_8xb16_cifar.py',
+    '../_base_/models/mobilenet_v3_small_cifar.py',
     '../_base_/datasets/cifar10_bs16.py',
     '../_base_/schedules/cifar10_bs128.py', '../_base_/default_runtime.py'
 ]
@@ -13,8 +13,3 @@ param_scheduler = dict(
 )
 
 train_cfg = dict(by_epoch=True, max_epochs=200)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (16 samples per GPU)
-auto_scale_lr = dict(base_batch_size=128)

configs/poolformer/poolformer-m36_32xb128_in1k.py

@@ -11,7 +11,7 @@ optim_wrapper = dict(
     clip_grad=dict(max_norm=5.0),
 )
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/poolformer/poolformer-m48_32xb128_in1k.py

@@ -11,7 +11,7 @@ optim_wrapper = dict(
     clip_grad=dict(max_norm=5.0),
 )
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/poolformer/poolformer-s12_32xb128_in1k.py

@@ -11,7 +11,7 @@ optim_wrapper = dict(
     clip_grad=dict(max_norm=5.0),
 )
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/poolformer/poolformer-s24_32xb128_in1k.py

@@ -11,7 +11,7 @@ optim_wrapper = dict(
     clip_grad=dict(max_norm=5.0),
 )
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/poolformer/poolformer-s36_32xb128_in1k.py

@@ -11,7 +11,7 @@ optim_wrapper = dict(
     clip_grad=dict(max_norm=5.0),
 )
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (32 GPUs) x (128 samples per GPU)
 auto_scale_lr = dict(base_batch_size=4096)

configs/regnet/regnetx-1.6gf_8xb128_in1k.py

@@ -4,8 +4,3 @@ _base_ = ['./regnetx-400mf_8xb128_in1k.py']
 model = dict(
     backbone=dict(type='RegNet', arch='regnetx_1.6gf'),
     head=dict(in_channels=912, ))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/regnet/regnetx-12gf_8xb64_in1k.py

@@ -12,7 +12,7 @@ train_dataloader = dict(batch_size=64)
 # for batch_size 512, use lr = 0.4
 optim_wrapper = dict(optimizer=dict(lr=0.4))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/regnet/regnetx-3.2gf_8xb64_in1k.py

@@ -12,7 +12,7 @@ train_dataloader = dict(batch_size=64)
 # for batch_size 512, use lr = 0.4
 optim_wrapper = dict(optimizer=dict(lr=0.4))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/regnet/regnetx-4.0gf_8xb64_in1k.py

@@ -12,7 +12,7 @@ train_dataloader = dict(batch_size=64)
 # for batch_size 512, use lr = 0.4
 optim_wrapper = dict(optimizer=dict(lr=0.4))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/regnet/regnetx-400mf_8xb128_in1k.py

@@ -34,7 +34,9 @@ train_pipeline = [
     dict(type='PackClsInputs'),
 ]
 
-train_dataloader = dict(dataset=dict(pipeline=train_pipeline))
+train_dataloader = dict(batch_size=128, dataset=dict(pipeline=train_pipeline))
+val_dataloader = dict(batch_size=128)
+test_dataloader = dict(batch_size=128)
 
 # schedule settings
 
@@ -53,8 +55,3 @@ custom_hooks = [
         interval=1,
         priority='ABOVE_NORMAL')
 ]
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/regnet/regnetx-6.4gf_8xb64_in1k.py

@@ -12,7 +12,7 @@ train_dataloader = dict(batch_size=64)
 # for batch_size 512, use lr = 0.4
 optim_wrapper = dict(optimizer=dict(lr=0.4))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/regnet/regnetx-8.0gf_8xb64_in1k.py

@@ -12,7 +12,7 @@ train_dataloader = dict(batch_size=64)
 # for batch_size 512, use lr = 0.4
 optim_wrapper = dict(optimizer=dict(lr=0.4))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/regnet/regnetx-800mf_8xb128_in1k.py

@@ -4,8 +4,3 @@ _base_ = ['./regnetx-400mf_8xb128_in1k.py']
 model = dict(
     backbone=dict(type='RegNet', arch='regnetx_800mf'),
     head=dict(in_channels=672, ))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (128 samples per GPU)
-auto_scale_lr = dict(base_batch_size=1024)

configs/repmlp/repmlp-base_8xb64_in1k-256px.py

@@ -30,7 +30,7 @@ test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 # schedule settings
 optim_wrapper = dict(clip_grad=dict(max_norm=1.0))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/repmlp/repmlp-base_8xb64_in1k.py

@@ -20,7 +20,7 @@ test_dataloader = dict(dataset=dict(pipeline=test_pipeline))
 # schedule settings
 optim_wrapper = dict(clip_grad=dict(max_norm=5.0))
 
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
 # base_batch_size = (8 GPUs) x (64 samples per GPU)
 auto_scale_lr = dict(base_batch_size=512)

configs/repmlp/repmlp-base_delopy_8xb64_in1k.py

@@ -1,8 +1,3 @@
 _base_ = ['./repmlp-base_8xb64_in1k.py']
 
 model = dict(backbone=dict(deploy=True))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=512)

configs/repmlp/repmlp-base_deploy_8xb64_in1k-256px.py

@@ -1,8 +1,3 @@
 _base_ = ['./repmlp-base_8xb64_in1k-256px.py']
 
 model = dict(backbone=dict(deploy=True))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (8 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=512)

configs/repvgg/repvgg-A0_4xb64-coslr-120e_in1k.py

@@ -10,8 +10,3 @@ param_scheduler = dict(
     type='CosineAnnealingLR', T_max=120, by_epoch=True, begin=0, end=120)
 
 train_cfg = dict(by_epoch=True, max_epochs=120)
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/repvgg/repvgg-A1_4xb64-coslr-120e_in1k.py

@@ -1,8 +1,3 @@
 _base_ = './repvgg-A0_4xb64-coslr-120e_in1k.py'
 
 model = dict(backbone=dict(arch='A1'))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)

configs/repvgg/repvgg-A2_4xb64-coslr-120e_in1k.py

@@ -1,8 +1,3 @@
 _base_ = './repvgg-A0_4xb64-coslr-120e_in1k.py'
 
 model = dict(backbone=dict(arch='A2'), head=dict(in_channels=1408))
-
-# NOTE: `auto_scale_lr` is for automatically scaling LR,
-# USER SHOULD NOT CHANGE ITS VALUES.
-# base_batch_size = (4 GPUs) x (64 samples per GPU)
-auto_scale_lr = dict(base_batch_size=256)