update README.md

README.md

 # MMPretrain-MMCV
+
+## 环境配置
+
+### Docker（方法一）
+
+推荐使用docker方式运行，拉取提供的docker镜像
+
+```shell
+docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-ubuntu20.04-dtk24.04.1-py3.10
+```
+
+基于拉取的镜像创建容器
+
+```shell
+# <your IMAGE ID or NAME>用以上拉取的docker的镜像ID或名称替换 
+docker run -it --name=mobilenetv2 --network=host --ipc=host --shm-size=16g --device=/dev/kfd --device=/dev/dri --device=/dev/mkfd --group-add video --privileged --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v /opt/hyhal:/opt/hyhal:ro -v $PWD/mobilenetv2_mmcv:/home/mobilenetv2_mmcv <your IMAGE ID> bash
+```
+
+克隆git仓库，并安装相关依赖
+
+```shell
+git clone --recursive http://developer.hpccube.com/codes/modelzoo/mobilenetv2_mmcv.git 
+cd mobilenetv2_mmcv/mmpretrain-mmcv 
+pip install -r requirements.txt
+```
+
+### Dockerfile（方法二）
+
+    cd mobilenetv2_mmcv/docker
+    docker build --no-cache -t mobilenetv2_mmcv:latest .
+    docker run -it --name=mobilenetv2 --network=host --ipc=host --shm-size=16g  --device=/dev/kfd --device=/dev/dri --device=/dev/mkfd --group-add video --privileged --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v /opt/hyhal:/opt/hyhal:ro -v $PWD/mobilenetv2_mmcv:/home/mobilenetv2_mmcv <your IMAGE ID> bash
+    # 若遇到Dockerfile启动的方式安装环境需要长时间等待，可注释掉里面的pip安装，启动容器后再安装python库：pip install -r requirements.txt
+
+### Anaconda（方法三）
+
+1、关于本项目DCU显卡所需的特殊深度学习库可从光合开发者社区下载安装： https://developer.hpccube.com/tool/
+
+```shell
+DTK驱动: DTK-24.04.1 
+python==3.10 
+torch==2.1.0 
+torchvision==0.16.0+das1.1.git7d45932.abi1.dtk2404.torch2.1 mmcv==2.0.1+das1.1.gite58da25.abi1.dtk2404.torch2.1.0
+```
+
+Tips：以上dtk驱动、python、torch等DCU相关工具版本需要严格一一对应
+
+2、其它非特殊库参照requirements.txt安装
+    pip install -r requirements.txt
+
+## 示例
+
+本仓库中提供了几个在tiny imagenet下进行测试的脚本
+
+如用8卡从零开始训练resnet50的运行方式如下，
+
+```shell
+bash tools/dist_train.sh resnet50-test.py 8
+```
+
+更多的配置在configs目录下，均可通过以下方式运行
+
+```shell
+bash tools/dist_train.sh <配置文件脚本> <训练用卡数>
+```
+
+

configs/_base_/models/tiny_mobilenet_v2_1x.py

+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='MobileNetV2', widen_factor=1.0),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=200,
+        in_channels=1280,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))

inception-v3_8xb32_in1k.py → inception-v3-test.py

@@ -5,11 +5,11 @@ _base_ = [
     'configs/_base_/default_runtime.py',
 ]
 
-import os
-import torch
+#import os
+#import torch
 
-torch.backends.cuda.matmul.allow_tf32=True
-torch.backends.cudnn.allow_tf32=True
+#torch.backends.cuda.matmul.allow_tf32=True
+#torch.backends.cudnn.allow_tf32=True
 
 train_pipeline = [
     dict(type='LoadImageFromFile'),
@@ -36,11 +36,11 @@ optim_wrapper = dict(
     optimizer=dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001))
 
 # 自定义hooks，添加ProfilerHook, 只在rank0启用
-custom_hooks = [
-    dict(type='ProfilerHook', by_epoch=False,
-        profile_times=5,
-        on_trace_ready=dict(type="log_trace", sort_by="self_cuda_time_total"),
-        json_trace_path=f"trace_inceptionv3_tf32.json",
-        activity_with_cuda=True,
-        schedule=dict(wait=3, warmup=1, active=1, repeat=1))  # 这样的设置是10次
-] if os.environ['LOCAL_RANK'] == '0' else []
+#custom_hooks = [
+#    dict(type='ProfilerHook', by_epoch=False,
+#        profile_times=5,
+#        on_trace_ready=dict(type="log_trace", sort_by="self_cuda_time_total"),
+#        json_trace_path=f"trace_inceptionv3_tf32.json",
+#        activity_with_cuda=True,
+#        schedule=dict(wait=3, warmup=1, active=1, repeat=1))  # 这样的设置是10次
+#] if os.environ['LOCAL_RANK'] == '0' else []

mobilenet-v2-test.py

+_base_ = [
+    'configs/_base_/models/tiny_mobilenet_v2_1x.py',
+    'configs/_base_/datasets/tiny_imagenet_bs32_pil_resize.py',
+    'configs/_base_/schedules/imagenet_bs256_epochstep.py',
+    'configs/_base_/default_runtime.py'
+]

resnet50-test.py

@@ -3,23 +3,20 @@ _base_ = [
     'configs/_base_/schedules/imagenet_bs256.py', 'configs/_base_/default_runtime.py'
 ]
 
-import torch
-
-torch.backends.cuda.matmul.allow_tf32=True
-torch.backends.cudnn.allow_tf32=True
+#import os
 
 # optimizer
 optim_wrapper = dict(
-        #type='AmpOptimWrapper',
-        #dtype='float16',
+        type='AmpOptimWrapper',
+        dtype='bfloat16',
     optimizer=dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001))
 
-custom_hooks = [
-        dict(type='ProfilerHook', by_epoch=False,
-            profile_times=12,
-            with_stack=True,
-            with_flops=True,
-            on_trace_ready=dict(type="log_trace", sort_by="self_cuda_time_total"),
-            activity_with_cuda=True,
-            schedule=dict(wait=1, warmup=1, active=10, repeat=1))
-        ]
+# 自定义hooks，添加ProfilerHook, 只在rank0启用
+#custom_hooks = [
+#    dict(type='ProfilerHook', by_epoch=False,
+#        profile_times=12,
+#        on_trace_ready=dict(type="log_trace", sort_by="self_cuda_time_total"),
+#        json_trace_path=f"trace_resnet50_8xb32_bf16.json",
+#        activity_with_cuda=True,
+#        schedule=dict(wait=1, warmup=1, active=10, repeat=1))  # 这样的设置是10次
+#] if os.environ['LOCAL_RANK'] == '0' else []

resnet50_imagenet200_8b32.py

-_base_ = [
-    'configs/_base_/models/resnet50.py', 'configs/_base_/datasets/tiny_imagenet_bs32.py',
-    'configs/_base_/schedules/imagenet_bs256.py', 'configs/_base_/default_runtime.py'
-]
-
-import os
-
-# optimizer
-optim_wrapper = dict(
-        type='AmpOptimWrapper',
-        dtype='bfloat16',
-    optimizer=dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001))
-
-# 自定义hooks，添加ProfilerHook, 只在rank0启用
-custom_hooks = [
-    dict(type='ProfilerHook', by_epoch=False,
-        profile_times=12,
-        on_trace_ready=dict(type="log_trace", sort_by="self_cuda_time_total"),
-        json_trace_path=f"trace_resnet50_8xb32_bf16.json",
-        activity_with_cuda=True,
-        schedule=dict(wait=1, warmup=1, active=10, repeat=1))  # 这样的设置是10次
-] if os.environ['LOCAL_RANK'] == '0' else []