LLMs streaming with AI Endpoints and LangChain4j

[News]

---

After explaining how to use AI Endpoints and LangChain4j in the previous post, let’s take a look at how to use streaming to create a real chat bot !

Create the project with Quarkus​

Like the previous post about AI Endpoints we’ll use LangChain4j through a Quarkus extension: quarkus-langchain4j.

All the code source used in the blog post is available on our GitHub repository: public-cloud-examples/ai/ai-endpoints/quarkus-langchain4j-streaming

First of all, you need to create a Quarkus application with the Quarkus CLI.

quarkus create app com.ovhcloud.examples.aiendpoints:quarkus-langchain4j-streaming \
    --extension='quarkus-langchain4j-mistral-ai,rest'

For more details on the type of files are created, see the blog post How to use AI Endpoints and LangChain4j

AI Service creation​

LLM is used remotely via a LangChain4j AI Service.

Let’s code our service to create a chat bot.

package com.ovhcloud.examples.aiendpoints.services;

import dev.langchain4j.service.SystemMessage;
import dev.langchain4j.service.UserMessage;
import io.quarkiverse.langchain4j.RegisterAiService;
import io.smallrye.mutiny.Multi;

@RegisterAiService
public interface ChatBotService {
  // Scope / context passed to the LLM
  @SystemMessage("You are a virtual, an AI assistant.")
  // Prompt (with detailed instructions and variable section) passed to the LLM
  @UserMessage("Answer as best possible to the following question: {question}. The answer must be in a style of a virtual assistant and add some emojis.")
  Multi<String> askAQuestion(String question);
}

If you want more information about the purpose of SystemMessage and UserMessage you can read the documentation of the Quarkus extension.

Then, you have to configure the quarkus-langchain4j extension to use AI Endpoints.

### Global configurations
# Base URL for Mistral AI endpoints
quarkus.langchain4j.mistralai.base-url=https://mistral-7b-instruct-v02.endpoints.kepler.ai.cloud.ovh.net/api/openai_compat/v1
# Activate or not the log during the request
quarkus.langchain4j.mistralai.log-requests=true
# Activate or not the log during the response
quarkus.langchain4j.mistralai.log-responses=true
# Delay before raising a timeout exception                   
quarkus.langchain4j.mistralai.timeout=60s   
# No key is needed
quarkus.langchain4j.mistralai.api-key=foo
 
# Activate or not the Mistral AI embedding model                     
quarkus.langchain4j.mistralai.embedding-model.enabled=false
 
### Chat model configurations
# Activate or not the Mistral AI chat model
quarkus.langchain4j.mistralai.chat-model.enabled=true             
# Chat model name used
quarkus.langchain4j.mistralai.chat-model.model-name=Mistral-7B-Instruct-v0.2
# Number of tokens to use
quarkus.langchain4j.mistralai.chat-model.max-tokens=1024

To know how to use the OVHcloud AI Endpoints product, please read the blog post: Enhance your applications with AI Endpoints

AI chat bot API​

Now it’s time to test our AI!

Let’s develop a small API.

package com.ovhcloud.examples.aiendpoints;

import com.ovhcloud.examples.aiendpoints.services.ChatBotService;
import io.smallrye.mutiny.Multi;
import jakarta.inject.Inject;
import jakarta.ws.rs.POST;
import jakarta.ws.rs.Path;

@Path("/ovhcloud-ai")
public class AIEndpointsResource {
    // AI Service injection to use it later
    @Inject
    ChatBotService chatBotService;

    // ask resource exposition with POST method
    @Path("ask")
    @POST
    public Multi<String> ask(String question) {
        // Call the Mistral AI chat model
        return chatBotService.askAQuestion(question);
    }
}

Call the API with a curl command.

curl -N http://localhost:8080/ovhcloud-ai/ask \                                                 
     -X POST \
     -d '{"question":"Can you tell me what OVHcloud is and what kind of products it offers?"}' \
     -H 'Content-Type: application/json'

And that it!

In just a few lines of code, you’ve developed your first chat bot using Quarkus, LangChain4j and AI Endpoints.

This chat bot use the streaming mode, fell free to develop a web application to be more usable by humans

.

Don’t hesitate to test our new product, AI Endpoints, and give us your feedback.

You have a dedicated Discord channel (#ai-endpoints) on our Discord server (https://discord.gg/ovhcloud), see you there!

Continue reading...