Skip to content

Quickly Build Dify LLM Application Development Platform

Dify is an open-source large language model (LLM) application development platform that provides one-stop capabilities including Agent workflow, RAG Pipeline, rich integrations, and observability, enabling users to quickly build production-grade generative AI applications.

This article mainly introduces how to use Helm Application in DCE to deploy dify-chart plugin, quickly build Dify LLM application development platform, and implement usage examples of Dify application/workflow based on Qwen-turbo.

Prerequisites

Before installing dify-chart plugin, the following prerequisites must be met:

Installation Process

Follow these steps to install dify-chart plugin for the cluster and build the Dify LLM application development platform.

  1. Find the target cluster where you want to install dify-chart plugin in the cluster list. Click the cluster name, then click Helm Applications -> Helm Templates in the left navigation bar. Enter dify-chart in the search bar.

    Cluster Details

  2. Read the dify-chart plugin introduction, select the version and click Install. This article uses 0.0.2 version as an example.

    Click Install

  3. Fill in and configure parameters, then click Next.

    Fill Parameters

    • Name: Required parameter, enter the plugin name. Note that the name can be at most 63 characters, can only contain lowercase letters, numbers and separators ("-"), and must start and end with lowercase letters or numbers, e.g., dify-chart.
    • Namespace: The namespace where the plugin is installed. You can choose an existing namespace or create a new one. For example, create dify namespace.
    • Version: Plugin version, e.g., 0.0.2.
    • Delete on Failure: Optional parameter. When enabled, it will default to enable installation waiting. If installation fails, it will delete installation-related resources.
    • Wait for Ready: Optional parameter. When enabled, it will wait for all associated resources under the application to be in ready state before marking the application installation as successful.
    • Detailed Logs: Optional parameter. When enabled, it will output detailed logs of the installation process.

    Note

    After enabling Wait for Ready and/or Delete on Failure, the application will take a relatively long time to be marked as Running .

    Configure Service Parameters

    • service :
      • type : The access method of the Dify application. Keep the default Nodeport node access.
      • nodePort : The access port, defaulting to 30000 .
      • nodeIP : The externally accessible node IP, defaulting to 127.0.0.1 , which needs to be replaced with the node IP of the current cluster.

  4. After confirming that the YAML is correct, click OK to complete the installation of the dify-chart plugin. The system will then automatically redirect to the Helm Applications list page. After a few minutes, refresh the page and you will see the application that was just installed. View Status

Quick Access

  1. You can access it directly through the configured nodeIP : nodePort , or click Services -> dify-nginx-nodeport -> External Access in the UI to access the built Dify platform.

    Service Access

  2. Enter the administrator initialization password to verify and enter the initialization page. The default administrator initialization password is password .

    Access Page

  3. Set the administrator account of the Dify application, which is used to create applications and manage LLM providers. Click Settings to complete the initialization.

    Administrator Account

  4. Log in with the administrator account set in the previous step to enter the Dify platform home page.

    Application Home Page

Usage Example

This article takes the quick building of a Chinese-English Translation application/workflow as an example to do a simple practice on the Dify platform built in the DCE target cluster.

Model Integration

  1. On the Dify platform home page, click Username in the upper right corner to enter the Settings page.

    Settings

  2. Click Settings -> Model Providers to enter the model list and select the required model.

    Model Providers

  3. Configure the basic parameters of the model and click Confirm to add the LLM service. This article takes Tongyi Qianwen as an example.

    Configure Parameters

    • Model Type: Dify classifies models into the following 4 categories according to usage scenarios: LLM Model, TTS Model, Text Embedding Model and Rerank Model.
    • Model Name: The specific name of the required model, e.g., qwen-turbo.
    • API Key: The core credential for calling the model service, used to verify user identity and protect data security.
    • Model Context Length: The maximum number of tokens (including input and output) that the model can retain when processing text at a time. Exceeding the context length will cause historical conversations to be discarded. Default: 4096.
    • Maximum Token Limit: The maximum number of tokens the model outputs at a time. Default: 4096.
    • Function Calling: Allows the model to call external tools or APIs when generating text. Default: Not Supported .

Build an Application

Steps to create a Text Generation application:

  1. On the Dify platform home page, click Create Application -> Create Blank Application , select the application type, and fill in the Application Name & Icon and Description to create the application.

    Create Application

  2. After the application is created, you will be automatically redirected to the application overview page. Click Orchestrate in the left menu to orchestrate the application. The Debug and Preview area on the right side of the interface allows you to debug and preview the application.

    Orchestrate Application

    • Prefix Prompt: Prompts are used to constrain AI to give professional replies, making the responses more precise. Users can use the built-in prompt generator to write suitable prompts.
    • Variables: Variables will be presented in the form of a form for users to fill in before the conversation. The values of the variables in the prompt will be replaced with the values filled in by the user. The maximum length can be set, for example, {{query}}.
    • Context: Context can be understood as the background information provided to the LLM, and is often used to fill in the output variables of knowledge retrieval.

  3. Click Publish in the upper right corner to complete the quick building of a simple application.

    Publish Application

  4. Click Publish -> Run in sequence to access the home page of the created application for use.

    Use Application

Workflow Configuration

Dify workflows are divided into two types:

  • Chatflow: Oriented to conversational scenarios, including customer service, semantic search, and other conversational applications that require multi-step logic when building responses.
  • Workflow: Oriented to automation and batch processing scenarios, suitable for applications such as high-quality translation, data analysis, content generation, and email automation.

Steps to create a Workflow workflow:

  1. On the Create Application -> Create Blank Application page, select Workflow, and fill in the Application Name & Icon and Description to create the workflow.

    Create Workflow

  2. After creation, you will be automatically redirected to the orchestration page. Add nodes by right-clicking or clicking the + sign at the end of the previous node to orchestrate the workflow.

    Orchestrate Workflow

  3. Add an LLM node, call the LLM, then fill in and configure the parameters.

    Call the LLM

    • Model: Call the LLM to answer questions, e.g., qwen-turbo.
    • Context: Context can be understood as the background information provided to the LLM, and is often used to fill in the output variables of knowledge retrieval.
    • Prompt: An easy-to-use prompt orchestration page. If you select a Chat model, you can customize the SYSTEM / USER / ASSISTANT parts.
    • Vision: Enabling the vision feature will allow the model to receive images as input and answer user questions based on the understanding of the image content.
    • Output Variables: The content generated by the LLM.
    • Error Retry: After enabling the error retry feature, the node will automatically retry according to a preset policy when an error occurs. You can adjust the maximum number of retries and the interval between retries to set the retry policy.
    • Exception Handling: Provides diversified node error handling policies, which can throw fault information when an error occurs at the current node without interrupting the main process; or continue to complete the task through an alternative path.

  4. Add the End node to complete the workflow orchestration. Click Run in the upper right corner to debug and preview the workflow.

    Debug Workflow

  5. Click Publish in the upper right corner to complete the quick building of a simple workflow.

    Call the LLM

For more features of the Dify application development platform, refer to Dify Official Documentation.

Comments