如何在运行时在tf_serving中添加多个模型而不卸载先前的模型? [英] How to add multiple models at runtime in tf_serving without unloading the previous models?

问题描述

我正在尝试使用以下代码在 tensorflow_model_server 中添加新模型:

I'm trying to add a new model in tensorflow_model_server using the following code:

from tensorflow_serving.apis import model_service_pb2_grpc
from tensorflow_serving.apis import model_management_pb2
from tensorflow_serving.config import model_server_config_pb2

import grpc

def add_model_config(host, name, base_path, model_platform):
  channel = grpc.insecure_channel(host)
  stub = model_service_pb2_grpc.ModelServiceStub(channel)
  request = model_management_pb2.ReloadConfigRequest()
  model_server_config = model_server_config_pb2.ModelServerConfig()

  #Create a config to add to the list of served models
  config_list = model_server_config_pb2.ModelConfigList()
  one_config = config_list.config.add()
  one_config.name= name
  one_config.base_path=base_path
  one_config.model_platform=model_platform

  model_server_config.model_config_list.CopyFrom(config_list)

  request.config.CopyFrom(model_server_config)

  print(request.IsInitialized())
  print(request.ListFields())

  response = stub.HandleReloadConfigRequest(request,10)
  if response.status.error_code == 0:
      print("Reload sucessfully")
      return True
  else:
      print("Reload failed!")
      print(response.status.error_code)
      print(response.status.error_message)
      return False

但是，每当我尝试执行它时，服务器就会卸载所有先前的模型.是否可以继续为所有模型提供服务，然后添加新模型?

But whenever I try to execute it the server unloads all of the previous models. Is it possible to keep serving all the models and then add a new model ?

推荐答案

没有简单的方法来添加新模型，同时保持先前模型已加载.

There is no easy way to just add a new model while keeping the previous ones already loaded.

我一直在做的是始终将最后一个模型配置列表保留在磁盘上，当出于任何原因(添加，删除或更新)需要刷新模型时，我都会从磁盘读取该配置文件，然后执行进行适当的修改，然后使用完整的配置列表调用HandleReloadConfigRequest()，然后再次将其保存到磁盘.

What I have been doing is to always keep the last model config list on disk, and when I need to refresh the models for any reason (either add or remove or update), I read that config file from disk, and do the proper modification, and call the HandleReloadConfigRequest() with the full config list, and then save it to disk again.

磁盘上的文件(例如/models/models.config )成为在任何给定时间已将哪些模型加载到tf.serve中的权威记录.这样，您就可以从tf.serve重新启动中恢复，并且可以轻松地知道它将加载正确的模型.在服务器启动期间用于指定配置文件的选项为-model_config_file/models/models.config.

The file on disk (say /models/models.config) becomes the authoritative record of what models have been loaded into tf.serve at any given time. This way you can recover from a tf.serve reboot and have the comfort knowing that it will load the correct models. The option for specifying the config file during server start is --model_config_file /models/models.config.

这篇关于如何在运行时在tf_serving中添加多个模型而不卸载先前的模型?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！