Document Upload in Open WebUI

EveryoneUpdated 11 Jun 2024LLM & Open WebUI
Document Upload in Open WebUI
Overview

This guide explains how to upload documents and configure the document upload feature in Open WebUI, including integration with GRACE. It covers both user and administrator perspectives, detailing how to upload documents, select models, and adjust embedding and extraction settings for optimal performance.

User Guide
For Normal Users
  1. Uploading Documents:
    • Click the plus (+) button in Open WebUI.
    • Select the document you wish to upload (e.g., PDF, DOCX, and other supported formats).
    • Once uploaded, the document will be processed using an embedding model and extraction mechanism to make its content available for queries.
  2. Using Uploaded Documents:
    • Choose your preferred model (e.g., OpenChatBT 4.1 with standard guardrails) from the model selection menu.
    • Enter your query or prompt (e.g., "Summarize this document").
    • The system will use the uploaded document as context to generate a response, applying guardrails if enabled.
For Administrators
  1. Accessing Document Upload Settings:
    • Navigate to the Admin Panel > Settings > Documents.
  2. Configuring Embedding Models:
    • By default, a local embedding model (e.g., Sentence Transformers) is used for document processing.
    • You can select a different Sentence Transformer model and download it if your system supports it.
    • For higher performance, select OpenAI as the embedding model and provide the endpoint URL for your deployed model on GRACE.
    • To use a custom embedding model hosted on GRACE:
      • Go to Connections in GRACE to find your model's endpoint URL.
      • In Open WebUI, set the embedding model to OpenAI and enter the GRACE endpoint URL.
      • Ensure the model name matches your deployment (e.g., "text-embedding-3-small" for Azure OpenAI).
  3. Adjusting Document Upload Parameters:
    • Configure the number of top hits extracted from vector storage to balance performance and relevance.
    • Customize the prompt template used to generate responses from uploaded documents. If left empty, the default template is used.
    • Set parameters for embedding creation, such as chunk size and overlap between document segments.
    • Limit the maximum file upload size (e.g., 20 MB) to prevent system overload. Leave empty for no limit.
  4. Important Note on Configuration Changes:
    • Changing embedding or extraction settings will affect all existing documents in the knowledge base.
    • After making significant changes, you must re-upload all documents to ensure consistency and optimal performance.
  5. Best Practices:
    • Refer to the official documentation for detailed configuration options and recommendations.
    • Test changes in a controlled environment before applying them to production systems.
Summary
  • Normal users can easily upload and query documents in Open WebUI.
  • Administrators have full control over embedding models, extraction settings, and upload limits, including integration with GRACE-hosted models.
  • Configuration changes may require re-uploading documents to maintain system accuracy and performance.