Serving: Is there an example of platform_config_file

Created on 3 Mar 2017 · 6Comments · Source: tensorflow/serving

When I want to start Tensorflow Serving with platform_config_file like:

 platform_configs: {
  key: "tensorflow",
  value:  {
          source_adapter_config: {
            config: {
              inter_op_parallelism_threads: 1
            }
          }
  }
}

and I got error like:

 [libprotobuf ERROR external/protobuf/src/google/protobuf/text_format.cc:299] Error parsing text-format tensorflow.serving.PlatformConfigMap: 5:19: Message type "google.protobuf.Any" has no field named "config".
2017-03-03 11:23:50.217710: F tensorflow_serving/model_servers/main.cc:267] Check failed: ::tensorflow::Status::OK() == (ParseProtoTextFile(file, &platform_config_map)) (OK vs. Invalid argument: Invalid protobuf file: './platform.conf')

Source

xiaop1987

👍4

Most helpful comment

based on the comment from @ales-t , I found more clear text-format

platform_configs {
  key: "tensorflow"
  value {
    source_adapter_config {
      [type.googleapis.com/tensorflow.serving.SavedModelBundleSourceAdapterConfig] {
        legacy_config {
          session_config {
            gpu_options {
              per_process_gpu_memory_fraction: 0.4
              allow_growth: true
            }
          }
        }
      }
    }
  }
}

you can easily change session_config part to fit your need.

to generate this file, use this script:

import tensorflow as tf

from tensorflow_serving.config import platform_config_pb2
from tensorflow_serving.servables.tensorflow import session_bundle_config_pb2
from tensorflow_serving.servables.tensorflow import saved_model_bundle_source_adapter_pb2


session_config = tf.ConfigProto()
# config whatever you want
session_config.gpu_options.allow_growth = True
session_config.gpu_options.per_process_gpu_memory_fraction = 0.4

legacy_config=session_bundle_config_pb2.SessionBundleConfig(session_config=session_config)
adapter = saved_model_bundle_source_adapter_pb2.SavedModelBundleSourceAdapterConfig(legacy_config=legacy_config)

config_map = platform_config_pb2.PlatformConfigMap()
config_map.platform_configs['tensorflow'].source_adapter_config.Pack(adapter)

print(config_map)

and to generate the *_pb2.py files, as follow:

# run at root of tensorflow_serving repo

TARGET_DIR="$1"

python -m grpc.tools.protoc \
    -I . -I ./tensorflow \
    --python_out "$TARGET_DIR" \
    tensorflow_serving/servables/tensorflow/saved_model_bundle_source_adapter.proto \
    tensorflow_serving/servables/tensorflow/session_bundle_config.proto \
    tensorflow_serving/config/platform_config.proto

pushd $TARGET_DIR

touch tensorflow_serving/__init__.py
touch tensorflow_serving/config/__init__.py
touch tensorflow_serving/servables/__init__.py
touch tensorflow_serving/servables/tensorflow/__init__.py

popd

also I put these scripts in this gist https://gist.github.com/cutewalker/58e1c4f71b5af4822bc732fd619ebda3

cutewalker on 17 Oct 2017

👍9 🎉2

All 6 comments

You can use something like this:
platform_configs {
key: "tensorflow"
value {
source_adapter_config {
type_url: "type.googleapis.com/tensorflow.serving.SavedModelBundleSourceAdapterConfig"
value: "..."
}
}
}

The "..." in the config should be replaced by the serialized SavedModelBundleSourceAdapterConfig that you want to use.

lilao on 17 Mar 2017

Have you solved this problem?I face the same problem. I want to write a platform_configs, allow_soft_placement.what should i do. thanks

baodingge on 1 Sep 2017

👍4

@baodingge

I faced pretty much the same problem and this is what I did:

install the protocol buffer compiler (latest version) by following the README:
https://github.com/google/protobuf/blob/master/src/README.md
clone TF serving repository and manually compile .proto files for Python:

~/protobuf/install/bin/protoc \
  -I=$(pwd) \
  -I=$(pwd)/tensorflow/ \
  --python_out=/tmp \
  $(pwd)/tensorflow_serving/servables/tensorflow/saved_model_bundle_source_adapter.proto
  $(pwd)/tensorflow_serving/servables/tensorflow/session_bundle_config.proto

install the generated *pb2.py files by copying them to my Python modules path, e.g.:
cp -r /tmp/tensorflow_serving $HOME/.local/lib/python3.5/site-packages/
write a Python script to generate the serialized protocol buffer which contains the required configuration:

#!/usr/bin/env python3

from tensorflow_serving.servables.tensorflow import session_bundle_config_pb2
from tensorflow_serving.servables.tensorflow import saved_model_bundle_source_adapter_pb2
from tensorflow.core.protobuf import config_pb2

adapter = saved_model_bundle_source_adapter_pb2.SavedModelBundleSourceAdapterConfig(
    legacy_config=session_bundle_config_pb2.SessionBundleConfig(
        session_config=config_pb2.ConfigProto(
            allow_soft_placement=True)))

print(adapter.SerializeToString())

the script outputs a serialized sequence of bytes, in this case "\xc2>\x04\x12\x028\x01"
you can put that in your platform config file, arriving finally at this:

platform_configs: {
  key: "tensorflow",
  value: {
    source_adapter_config: {
      type_url: "type.googleapis.com/tensorflow.serving.SavedModelBundleSourceAdapterConfig",
      value: "\xc2>\x04\x12\x028\x01"
    }
  }
}

Personally, I don't think that's a very intuitive way to configure anything but it does seem to work (for now).

ales-t on 5 Oct 2017

❤1