This document provides instructions for deploying Youtu Embedding, HiChunk, and Youtu Parsing backend services on TencentCloud Ti-one platform.

Overview

TencentCloud Ti-one is a one-stop machine learning platform that provides model training, deployment, and inference services. Using Ti-one, you can deploy the Youtu-RAG backend services without managing your own GPU infrastructure.

Prerequisites

• A TencentCloud account with Ti-one and Tencent Container Registry (TCR) access

Deploying Services on Ti-one

Step 1: Prepare Container Images

1. Build Docker images following the respective Docker deployment guides for each backend service
   • Youtu Embedding
   • Youtu HiChunk
   • Youtu Parsing
2. Upload the Docker images to TencentCloud Container Registry (TCR). You can follow the TCR documentation for instructions on pushing images to TCR.

Step 2: Create a Model Service

1. Log in to the TencentCloud Console and navigate to Ti-one
2. Go to Model Services > Online Services
3. Click Create Service

Step 3: Configure the Model

1. Service Name: Enter a name for your service (e.g., youtu-embedding)
2. Deployment Method: Select Standard deployment
3. Source of Machine: Choose Select from CVMs if you have existing instances, or Purchased on TI-ONE to create new instances
4. Model Source: Select Container Images and choose the container image you just uploaded to TencentCloud Container Registry (TCR)
5. Port: Default to 8501, or specify a custom port matching your container configuration
6. Spec: Choose a GPU instance

Step 4: Update the Service Endpoints

1. After creating the service, navigate to the service details page
2. Find and copy the Call Address under Regular Service Calling in the Service Call section. This will be your service endpoint for Youtu-RAG integration.

Configuring Youtu-RAG

Once all services are deployed on Ti-one, update your .env file with the Ti-one service endpoints. You will also need to check your RAG configuration files as indicated in the local deployment guide to ensure compatibility.

# =============================================
# Embeddings (Ti-one)
# =============================================
UTU_EMBEDDING_URL=https://<your-tione-embedding-endpoint>
UTU_EMBEDDING_MODEL=youtu-embedding-2B

# =============================================
# Chunk (Ti-one)
# =============================================
UTU_CHUNK_BASE_URL=https://<your-tione-hichunk-endpoint>
UTU_CHUNK_MODEL=hichunk

# =============================================
# OCR (Ti-one)
# =============================================
UTU_OCR_BASE_URL=https://<your-tione-parsing-endpoint>
UTU_OCR_MODEL=youtu-parsing

Additional Resources

• TencentCloud Ti-one Documentation
• Ti-one Service Deployment Guide
• Ti-one Container Guide
• Ti-one Pricing