CODEWITHSUNDEEP
tensorflowtensorflow2.0tensorflow-serving
Sergiodiaz53
Sergiodiaz53

Reputation: 1288

Can't setup specific parameters of server_core through model_config_list

Im having an infinite loop when model_paths are not correct in model_config_list for tf-serving. This seems to be a known behaviour for tf-serving and it also seems to be two different ways of avoiding it.

Both of them are related to two different options that are setup in the server_core.c. First one is servable_versions_always_present which is present in my current version (2.14.1) and the other one is should_retry_model_load which is in 2.18. The problem is that neither of them are exposed to be setup as a flag in the call (as you can see in the main.cc) so you can't just do tensorflow_model_server --model_config_file=/tf/models/models.config --servable_versions_always_present=true, for example.

The thing is that, even is not documented anywhere I could find, I saw a couple of comments in the github that recommend to setup this options through the models.config file but every try I have done has failed.

i.e:

model_config_list {
  config {
    name: "my_model"
    base_path: "/models/my_model/model"
    model_platform: "tensorflow"
}
servable_versions_always_present: true

or

model_config_list {
  config {
    name: "my_model"
    base_path: "/models/my_model/model"
    model_platform: "tensorflow"
    servable_versions_always_present: true
}

So the question is: is there anyway to setup any of those two parameters from the model.config or any other way without adding them in the main.cc as a flag or modifying them directly in the code?

Thanks in advance.

Upvotes: 0

Views: 18

Answers (0)

Related Questions