watsonx-go

watsonx-go is a watsonx Client for Go

Install

go get -u github.com/IBM/watsonx-go

Usage

import (
  wx "github.com/IBM/watsonx-go/pkg/models"
)

Example Usage

export WATSONX_API_KEY="YOUR WATSONX API KEY"
export WATSONX_PROJECT_ID="YOUR WATSONX PROJECT ID"

Create a client:

client, _ := wx.NewClient()

Or pass in the required secrets directly:

client, err := wx.NewClient(
  wx.WithWatsonxAPIKey(apiKey),
  wx.WithWatsonxProjectID(projectID),
)

Generate Text

Generation:

result, _ := client.GenerateText(
  "meta-llama/llama-3-1-8b-instruct",
  "Hi, who are you?",
  wx.WithTemperature(0.4),
  wx.WithMaxNewTokens(512),
)

println(result.Text)

Stream Generation:

dataChan, _ := client.GenerateTextStream(
  "meta-llama/llama-3-1-8b-instruct",
  "Hi, who are you?",
  wx.WithTemperature(0.4),
  wx.WithMaxNewTokens(512),
)

for data := range dataChan {
  print(data.Text) // print the result as it's being generated
}

Chat Completions

Simple chat with a single message:

response, _ := client.SimpleChat(
  "meta-llama/llama-3-3-70b-instruct",
  "What is the capital of France?",
  wx.WithChatTemperature(0.3),
  wx.WithChatMaxTokens(100),
)

println(response) // prints the assistant's response text

Multi-turn conversation:

messages := []wx.ChatMessage{
  wx.CreateSystemMessage("You are a helpful assistant."),
  wx.CreateUserMessage("What is the capital of France?"),
}

response, _ := client.Chat(
  "meta-llama/llama-3-3-70b-instruct",
  messages,
  wx.WithChatTemperature(0.3),
  wx.WithChatMaxTokens(100),
)

content := response.Choices[0].Message.Content.GetText()
println(content)

Chat with JSON mode:

messages := []wx.ChatMessage{
  wx.CreateSystemMessage("You respond in JSON format."),
  wx.CreateUserMessage("What is 2+2? Respond with {\"answer\": number}"),
}

response, _ := client.Chat(
  "meta-llama/llama-3-3-70b-instruct",
  messages,
  wx.WithChatJSONMode(),
  wx.WithChatTemperature(0.1),
)

jsonResponse := response.Choices[0].Message.Content.GetText()
println(jsonResponse) // {"answer": 4}

Generate Embeddings

Embedding | Single query:

result, _ := client.EmbedQuery(
	"ibm/slate-30m-english-rtrvr",
	"Hello, world!",
	wx.WithEmbeddingTruncateInputTokens(2), 
	wx.WithEmbeddingReturnOptions(true),
)

embeddingVector := result.Results[0].Embedding

Embedding | Multiple docs:

result, _ := clientl.EmbedDocuments(
    "ibm/slate-30m-english-rtrvr",
    []string{"Hello, world!", "Goodbye, world!"},
    wx.WithEmbeddingTruncateInputTokens(2), 
    wx.WithEmbeddingReturnOptions(true),
)

for _, doc := range result.Results {
    fmt.Println(doc.Embedding)
}

Development Setup

Tests

Setup

export WATSONX_API_KEY="YOUR WATSONX API KEY"
export WATSONX_PROJECT_ID="YOUR WATSONX PROJECT ID"

Run

go test ./...

Pre-commit Hooks

Run the following command to run pre-commit formatting:

git config --local core.hooksPath .githooks/

Using Test Environment

There are two methods for configuring the watsonx client to be used with the test environment:

Option 1: Using Environment Variables

Specify the Watsonx URL and IAM endpoint using environment variables:

export WATSONX_URL_HOST="us-south.ml.test.cloud.ibm.com"
export WATSONX_IAM_HOST="iam.test.cloud.ibm.com"

Option 2: Using the NewClient Function Parameters

Specify the Watsonx URL and IAM endpoint through the parameters of the NewClient function:

client, err := wx.NewClient(
  wx.WithURL("us-south.ml.test.cloud.ibm.com"),
  wx.WithIAM("iam.test.cloud.ibm.com"),
  wx.WithWatsonxAPIKey(apiKey),
  wx.WithWatsonxProjectID(projectID),
)

---

Resources

• watsonx REST API Docs
• watsonx Python SDK Docs