Skip to content

Latest commit

 

History

History
280 lines (202 loc) · 9.94 KB

README.md

File metadata and controls

280 lines (202 loc) · 9.94 KB
Raw

Golang Install Guide

Documentation for installing the Go bindings for TensorFlow.

Maintainer: @wamuir

About

TensorFlow provides a Go API particularly useful for loading models created with Python and running them within a Go application.

Important: TensorFlow for Go is no longer supported by the TensorFlow team. The TensorFlow Go API is also not covered by the TensorFlow API stability guarantees.

Supported Platforms

The Go bindings for TensorFlow work on the following systems, and likely others:

  • Linux, 64-bit, x86
  • macOS, Version 10.12.6 (Sierra) or higher

Installation

Install the TensorFlow Go API from a community build (compatible with go get) or from source.

Community Build

Note: the Go bindings depend on libtensorflow, which should be downloaded (or compiled) and installed first.

A community build, Graft, contains nightly and release builds of the Go language bindings to the TensorFlow C API, including Go-compiled TensorFlow protocol buffers and generated Go wrappers for TensorFlow operations. This community build can be fetched using go get.

After installing libtensorflow, use Graft exactly as you would use the Go bindings found in the main TensorFlow repo, and with one of the following import statements:

TensorFlow C API Graft
TensorFlow Release 2.17.0 go get github.com/wamuir/graft/[email protected]
TensorFlow Release 2.16.2 go get github.com/wamuir/graft/[email protected]
TensorFlow Nightly go get github.com/wamuir/graft/tensorflow@nightly

Build from Source

Click to expand

Note: these build instructions are specific to TensorFlow 2.17.0

1. Install the TensorFlow C Library

Install the TensorFlow C library. This library is required for use of the TensorFlow Go package at runtime. For example, on Linux (64-bit, x86):

$ curl -L https://storage.googleapis.com/tensorflow/versions/2.17.0/libtensorflow-cpu-linux-x86_64.tar.gz | tar xz --directory /usr/local
$ ln -s external/local_tsl/tsl /usr/local/include/tsl
$ ldconfig

2. Install the Protocol Buffers Library and Compiler

Install the protocol buffers library and compiler. The compiler and well-known proto type files from the library are required during installation of the Go bindings.

  • Linux, using apt or apt-get, for example:

    $ apt install libprotobuf-dev protobuf-compiler

  • MacOS, using Homebrew:

    $ brew install protobuf

3. Install and Setup the TensorFlow Go API

The use of go get is not currently supported for installation of the TensorFlow Go API. Instead, follow these instructions.

  • First, note the location of your Go workspace. The remaining installation steps must be performed inside your Go workspace.

    $ go env GOPATH

  • Clone the TensorFlow source repository, substituting the location of your Go workspace for /go in the command below.

    $ git clone --branch v2.17.0 https://github.com/tensorflow/tensorflow.git /go/src/github.com/tensorflow/tensorflow

  • Change the working directory to the base of the cloned TensorFlow repository, substituting the the location of your Go workspace for /go in the command below.

    $ cd /go/src/github.com/tensorflow/tensorflow

  • Initialize a new go.mod file.

    $ go mod init github.com/tensorflow/tensorflow

  • Patch TensorFlow core protos to declare Go package.

    $ sed -i '4 i option go_package = "github.com\/tensorflow\/tensorflow\/tensorflow\/go\/core\/framework\/dataset_go_proto";' tensorflow/core/framework/dataset.proto
$ sed -i '9 c option go_package = "github.com\/tensorflow\/tensorflow\/tensorflow\/go\/core\/framework\/graph_debug_info_go_proto";' tensorflow/core/framework/graph_debug_info.proto
$ sed -i '4 i option go_package = "github.com\/tensorflow\/tensorflow\/tensorflow\/go\/core\/framework\/optimized_function_graph_go_proto";' tensorflow/core/framework/optimized_function_graph.proto

  • Patch Tensor Standard Library (TSL) protos to declare Go package.

    $ sed -i '5 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/bfc_memory_map.proto
$ sed -i '5 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/coordination_config.proto
$ sed -i '7 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/coordination_service.proto
$ sed -i '6 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/distributed_runtime_payloads.proto
$ sed -i '8 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/dnn.proto
$ sed -i '12 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/error_codes.proto
$ sed -i '8 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/histogram.proto
$ sed -i '5 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/rpc_options.proto
$ sed -i '10 c option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/status.proto
$ sed -i '13 i option go_package = "github.com\/google\/tsl\/tsl\/go_proto";' third_party/xla/third_party/tsl/tsl/protobuf/test_log.proto

  • Patch tensorflow/go/genop to generate TF and TSL protobufs.

    $ sed -i '71d;72d' tensorflow/go/genop/generate.sh
$ sed -i '71 i \    ${TF_DIR}\/third_party\/xla\/xla\/autotuning.proto \\' tensorflow/go/genop/generate.sh
$ sed -i '72 i \    ${TF_DIR}\/third_party\/xla\/third_party\/tsl\/tsl\/protobuf\/*.proto; do \\' tensorflow/go/genop/generate.sh
$ sed -i '74 i \    -I ${TF_DIR}/third_party/xla/third_party/tsl \\' tensorflow/go/genop/generate.sh
$ sed -i '75 i \    -I ${TF_DIR}/third_party/xla \\' tensorflow/go/genop/generate.sh

  • Generate wrappers and protocol buffers.

    $ (cd tensorflow/go/op && go generate)

  • Use Go Mod's replace directive to locate TSL protos.

    $ go mod edit -require github.com/google/[email protected]+incompatible
$ go mod edit -replace github.com/google/tsl=/go/src/github.com/google/tsl

  • Initialize a new go.mod for TSL and add dependencies.

    $ (cd /go/src/github.com/google/tsl && go mod init github.com/google/tsl && go mod tidy)

  • Add missing modules.

    $ go mod tidy

  • Test the installation.

    $ go test ./...

Usage

Applications must use Go Mod's replace directive

The replace directive instructs Go to use the local installation and must be added to go.mod for every Go module that depends on the API. Point the replace directive to the location within your Go workspace where you installed the API, substituting the location of your Go workspace for /go in the command below:

$ go mod init hello-world
$ go mod edit -require github.com/google/[email protected]+incompatible
$ go mod edit -require github.com/tensorflow/[email protected]+incompatible
$ go mod edit -replace github.com/google/tsl=/go/src/github.com/google/tsl
$ go mod edit -replace github.com/tensorflow/tensorflow=/go/src/github.com/tensorflow/tensorflow
$ go mod tidy

Example program

With the TensorFlow Go API installed, create an example program with the following source code (hello_tf.go):

package main

import (
	tf "github.com/tensorflow/tensorflow/tensorflow/go"
	"github.com/tensorflow/tensorflow/tensorflow/go/op"
	"fmt"
)

func main() {
	// Construct a graph with an operation that produces a string constant.
	s := op.NewScope()
	c := op.Const(s, "Hello from TensorFlow version " + tf.Version())
	graph, err := s.Finalize()
	if err != nil {
		panic(err)
	}

	// Execute the graph in a session.
	sess, err := tf.NewSession(graph, nil)
	if err != nil {
		panic(err)
	}
	output, err := sess.Run(nil, []tf.Output{c}, nil)
	if err != nil {
		panic(err)
	}
	fmt.Println(output[0].Value())
}

Initialize go.mod for the example program:

$ go mod init app
$ go mod edit -require github.com/google/[email protected]+incompatible
$ go mod edit -require github.com/tensorflow/[email protected]+incompatible
$ go mod edit -replace github.com/google/tsl=/go/src/github.com/google/tsl
$ go mod edit -replace github.com/tensorflow/tensorflow=/go/src/github.com/tensorflow/tensorflow
$ go mod tidy

Then, run the example program:

$ go run hello_tf.go

The command outputs: Hello from TensorFlow version *number*

Success: TensorFlow for Go has been configured.

Docker Example

A Dockerfile is available, which executes the installation and setup process for the Go bindings and builds the example program. To use, install Docker and then run the following commands:

$ docker build -t tensorflow/build:golang-example https://github.com/tensorflow/build.git#:golang_install_guide/example-program
$ docker run tensorflow/build:golang-example