添加openmmlab测试用例

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs2048.py

+# optimizer
+optimizer = dict(
+    type='SGD', lr=0.8, momentum=0.9, weight_decay=0.0001, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=2500,
+    warmup_ratio=0.25,
+    step=[30, 60, 90])
+runner = dict(type='EpochBasedRunner', max_epochs=100)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs2048_AdamW.py

+# optimizer
+# In ClassyVision, the lr is set to 0.003 for bs4096.
+# In this implementation(bs2048), lr = 0.003 / 4096 * (32bs * 64gpus) = 0.0015
+optimizer = dict(type='AdamW', lr=0.0015, weight_decay=0.3)
+optimizer_config = dict(grad_clip=dict(max_norm=1.0))
+
+# specific to vit pretrain
+paramwise_cfg = dict(
+    custom_keys={
+        '.backbone.cls_token': dict(decay_mult=0.0),
+        '.backbone.pos_embed': dict(decay_mult=0.0)
+    })
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    min_lr=0,
+    warmup='linear',
+    warmup_iters=10000,
+    warmup_ratio=1e-4)
+runner = dict(type='EpochBasedRunner', max_epochs=300)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs2048_coslr.py

+# optimizer
+optimizer = dict(
+    type='SGD', lr=0.8, momentum=0.9, weight_decay=0.0001, nesterov=True)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    min_lr=0,
+    warmup='linear',
+    warmup_iters=2500,
+    warmup_ratio=0.25)
+runner = dict(type='EpochBasedRunner', max_epochs=100)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs256.py

+# optimizer
+optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(policy='step', step=[30, 60, 90])
+runner = dict(type='EpochBasedRunner', max_epochs=100)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs256_140e.py

+# optimizer
+optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(policy='step', step=[40, 80, 120])
+runner = dict(type='EpochBasedRunner', max_epochs=140)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs256_coslr.py

+# optimizer
+optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(policy='CosineAnnealing', min_lr=0)
+runner = dict(type='EpochBasedRunner', max_epochs=100)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs256_epochstep.py

+# optimizer
+optimizer = dict(type='SGD', lr=0.045, momentum=0.9, weight_decay=0.00004)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+lr_config = dict(policy='step', gamma=0.98, step=1)
+runner = dict(type='EpochBasedRunner', max_epochs=300)

openmmlab_test/mmclassification-speed-benchmark/configs/_base_/schedules/imagenet_bs4096_AdamW.py

+# optimizer
+optimizer = dict(type='AdamW', lr=0.003, weight_decay=0.3)
+optimizer_config = dict(grad_clip=dict(max_norm=1.0))
+
+# specific to vit pretrain
+paramwise_cfg = dict(
+    custom_keys={
+        '.backbone.cls_token': dict(decay_mult=0.0),
+        '.backbone.pos_embed': dict(decay_mult=0.0)
+    })
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    min_lr=0,
+    warmup='linear',
+    warmup_iters=10000,
+    warmup_ratio=1e-4)
+runner = dict(type='EpochBasedRunner', max_epochs=300)

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/README.md

+# Mixed Precision Training
+
+## Introduction
+
+<!-- [OTHERS] -->
+
+```latex
+@article{micikevicius2017mixed,
+  title={Mixed precision training},
+  author={Micikevicius, Paulius and Narang, Sharan and Alben, Jonah and Diamos, Gregory and Elsen, Erich and Garcia, David and Ginsburg, Boris and Houston, Michael and Kuchaiev, Oleksii and Venkatesh, Ganesh and others},
+  journal={arXiv preprint arXiv:1710.03740},
+  year={2017}
+}
+```
+
+## Results and models
+
+|         Model         | Params(M) | Flops(G) | Mem (GB) | Top-1 (%) | Top-5 (%) | Config | Download |
+|:---------------------:|:---------:|:--------:|:---------:|:---------:|:---------:| :---------:|:--------:|
+| ResNet-50             | 25.56     | 4.12     |    1.9    |76.32 | 93.04 | [config](https://github.com/open-mmlab/mmclassification/blob/master/configs/fp16/resnet50_b32x8_fp16_imagenet.py) | [model](https://download.openmmlab.com/mmclassification/v0/fp16/resnet50_batch256_fp16_imagenet_20210320-b3964210.pth) &#124; [log](https://download.openmmlab.com/mmclassification/v0/fp16/resnet50_batch256_fp16_imagenet_20210320-b3964210.log.json) |

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/metafile.yml

+Collections:
+  - Name: FP16
+    Metadata:
+      Training Data: ImageNet
+      Training Resources: 8x V100 GPUs
+      Training Techniques:
+        - SGD with Momentum
+        - Weight Decay
+        - Mixed Precision Training
+    Paper: https://arxiv.org/abs/1710.03740
+    README: configs/fp16/README.md
+
+Models:
+- Config: configs/fp16/resnet50_b32x8_fp16_dynamic_imagenet.py
+  In Collection: FP16
+  Metadata:
+    FLOPs: 4120000000
+    Parameters: 25560000
+    Epochs: 100
+    Batch Size: 256
+    Architecture:
+      - ResNet
+  Name: resnet50_b32x8_fp16_dynamic_imagenet
+  Results:
+  - Dataset: ImageNet
+    Metrics:
+      Top 1 Accuracy: 76.32
+      Top 5 Accuracy: 93.04
+    Task: Image Classification
+  Weights: https://download.openmmlab.com/mmclassification/v0/fp16/resnet50_batch256_fp16_imagenet_20210320-b3964210.pth

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/resnet152_b32x8_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/resnet152.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py', '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/resnet18_b32x8_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/resnet18.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py', '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/resnet34_b32x8_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/resnet34.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py', '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/resnet50_b32x8_fp16_dynamic_imagenet.py

+_base_ = ['../resnet/resnet50_b32x8_imagenet.py']
+
+# fp16 settings
+fp16 = dict(loss_scale='dynamic')

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/resnet50_b32x8_fp16_imagenet.py

+#_base_ = ['../resnet/resnet50_b32x8_imagenet.py']
+
+_base_ = [
+    '../_base_/models/resnet50.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py', '../_base_/default_runtime.py'
+]
+
+
+# fp16 settings
+fp16 = dict(loss_scale=512.)

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/resnext50_32x4d_b32x8_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/resnext50_32x4d.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py', '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/seresnet50_b32x8_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/seresnet50.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256_140e.py',
+    '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/shufflenet_v1_1x_b64x16_linearlr_bn_nowd_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/shufflenet_v1_1x.py', '../speed_test/datasets/imagenet_bs64.py',
+    '../_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py',
+    '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/shufflenet_v2_1x_b64x16_linearlr_bn_nowd_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/shufflenet_v2_1x.py', '../speed_test/datasets/imagenet_bs64.py',
+    '../_base_/schedules/imagenet_bs1024_linearlr_bn_nowd.py',
+    '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file

openmmlab_test/mmclassification-speed-benchmark/configs/fp16/vgg11_b32x8_fp16_imagenet.py

+_base_ = [
+    '../_base_/models/vgg11.py', '../speed_test/datasets/imagenet_bs32.py',
+    '../_base_/schedules/imagenet_bs256.py', '../_base_/default_runtime.py'
+]
+# fp16 settings
+fp16 = dict(loss_scale=512.)
\ No newline at end of file