使用 Notebook 完成图像分类模型构建

from tikit.client import Client

# 初始化client。在公有云TI平台的环境（包括notebook和训练任务）中可以不填地区，环境变量中已经有地区信息。
# secret_id获取地址：https://console.cloud.tencent.com/cam/capi。
client = Client("your_secret_id", "your_secret_key", "<region>")

# 上传数据集（注：重复上传不会覆盖，会生成新的子目录，如有需要，可使用 client.delete_cos_path 删除旧数据）
client.upload_to_cos("LOCAL_PATH", "YOUR_BUCKE_NAME", "YOUR_COS_PATH")

--nnodes：对应环境变量NODE_NUM
--node_rank：对应环境变量INDEX
--nproc_per_node：对应环境变量GPU_NUM_PER_NODE
--master_addr：对应环境变量CHIEF_IP
--master_port：没有对应环境变量，用户自己指定一个即可


from tikit.client import Client
client = Client("your_secret_id", "your_secret_key", "<region>")

# 查看上传指标方法
help(client.push_training_metrics)


def validate(val_loader, model, criterion, epoch, args):
    batch_time = AverageMeter('Time', ':6.3f')
    losses = AverageMeter('Loss', ':.4e')
    top1 = AverageMeter('Acc@1', ':6.2f')
    top5 = AverageMeter('Acc@5', ':6.2f')
    progress = ProgressMeter(
        len(val_loader),
        [batch_time, losses, top1, top5],
        prefix='Test: ')

    # switch to evaluate mode
    model.eval()

    with torch.no_grad():
        end = time.time()
        for i, (images, target) in enumerate(val_loader):
            if args.gpu is not None:
                images = images.cuda(args.gpu, non_blocking=True)
            target = target.cuda(args.gpu, non_blocking=True)

            # compute output
            output = model(images)
            loss = criterion(output, target)

            # measure accuracy and record loss
            acc1, acc5 = accuracy(output, target, topk=(1, 5))
            losses.update(loss.item(), images.size(0))
            top1.update(acc1[0], images.size(0))
            top5.update(acc5[0], images.size(0))

            # measure elapsed time
            batch_time.update(time.time() - end)
            end = time.time()

            if i % args.print_freq == 0:
                progress.display(i)

        # 上传指标 
        client.push_training_metrics(int(time.time()), {"acc1": float(format(top1.avg, '.3f')), "acc5": float(format(top5.avg, '.3f'))}, epoch=epoch)
        # TODO: this should also be done with the ProgressMeter
        print('TIACC - * Acc@1 {top1.avg:.3f} Acc@5 {top5.avg:.3f} Epoch={epoch}'
              .format(top1=top1, top5=top5, epoch=epoch))

    return top1.avg

client.upload_to_cos("LOCAL_PATH", "YOUR_BUCKE_NAME", "YOUR_COS_PATH")

cd ~/classify_example_2022-04-28/code
source activate pytorch_py3
sh start_local.sh

# 查询平台支持的训练框架
client.describe_training_frameworks()

# 查询计算资源配置
client.describe_postpaid_training_price()

# 创建训练任务
from tikit import models

# 配置训练任务的计算框架和训练模式，此处选择训练框架为torch1.9-py3.8-cuda11.1-gpu，训练模式为DDP
framework = models.FrameworkInfo.new_system_framework("PYTORCH", "torch1.9-py3.8-cuda11.1-gpu", "DDP")
# 配置计算资源，此处配置为V100节点*2
worker_resource = models.ResourceConfigInfo.new_postpaid("TI.GN10.2XLARGE40.POST", 2)

# 配置数据输入信息，需要将COS中的训练数据映射到容器目录中
# 'demo-1256580188/notebook-demo-classification/data'是第一步数据上传的cos路径，
# '/opt/ml/input/data/image_classify'是步骤2.3中指定的数据读取路径，也就是容器本地路径（代码中读取）

input_data = models.TrainingDataConfig.new_cos_data({"demo-1256580188/notebook-demo-classification/data/":"/opt/ml/input/data/image_classify"})

# 创建训练任务
client.create_training_task("xxx5", # 任务名称
                            framework, # 训练框架
                            "demo-1256580188/output/", # 训练输出路径
                            worker_resource, # 资源配置信息
                            input_data_config=input_data, # 数据输入配置
                            code_package_path="demo-1256580188/notebook-demo-classification/code/", # 训练代码所在路径
                            worker_start_cmd="sh start_cloud.sh") # 启动命令



# 查看训练日志
client.describe_train_logs("train-544168070598700544*")


# 查看推理镜像。（运行查看最新版本信息）
client.describe_system_reasoning_images()

```

确定好推理框架后开始创建模型。

```
# 指定模型推理框架
reasoning_env = models.ReasoningEnvironment.new_system_environment("torch1.9.0-py38-cu111")
# 模型输出路径,即/model文件目录，为用户在训练任务配置的模型输出路径+task_id，如下output/train-544168070598700544/所示
model_output_path = models.CosPathInfo("demo-1256580188", "output/train-544168070598700544/", "ap-guangzhou")
# 创建模型，这里训练任务ID就是上面创建的训练任务ID
client.create_model_by_task("model1", "train-544168070598700544", reasoning_env, module_output_path, "PYTORCH")
```

模型创建成功后可在模型仓库查看对应的模型信息。