Getting Started with h2oGPTe: A Guide to Enterprise RAG
Science & Technology
Introduction
Welcome to H2O's premium AI Cloud environment! In this guide, we'll walk you through our Enterprise Retrieval-Augmented Generation (RAG) product, H2O GPTe. We'll start by explaining how to access the free trial environment so you can explore its features yourself.
Accessing the Free Trial Environment
To get started, visit the H2O website at h2o.ai. Here, you can find the various applications available, but our focus today is on the Enterprise H2O GPTe platform.
Logging In
To access the platform, you will need to log in with your personal Gmail or GitHub account. This step is necessary to prevent unauthorized usage of the system. For demonstration, I’ll log in using my GitHub account.
Building a Data Collection
Once logged in, you'll have access to the environment where you can chat with large language models based on specific data. Begin by clicking the edit icon at the bottom to create a new data collection.
Configuring Your Collection
You can customize your collection with the following settings:
- Name: Give your collection a suitable name (e.g., H2O Website).
- Embedding Model: Choose the embedding model to use (e.g., BG M3 for multilingual data).
- Token Settings: You can adjust chunk size and overlap for tokens; however, the default settings suffice for most users.
If your data includes private customer information, you can enable guardrails for private data detection. This feature allows you to redact sensitive information before incorporating it into public large models like OpenAI's GPT-4.
Adding Data to Your Collection
Once your collection is created, you can add data by uploading files from your local machine (like PDFs, audio files, or images) or by crawling a website. In our example, we'll crawl the H2O website.
You can specify that you only want to capture certain data without following links to other pages.
Data Processing Features
Automatic Summarization and Question Generation
For the documents added, H2O GPTe offers the option to automatically summarize the documents and generate sample questions for further exploration.
Support for Various Content Types
You can upload PDFs, URLs, audio files, and text files. The platform also supports Optical Character Recognition (OCR) for extracting text from images within documents.
Once the data is added, it will be embedded in a vector database for easy retrieval during chat interactions.
Querying the Data
To demonstrate H2O GPTe's capabilities, you can ask questions based on the data you’ve embedded.
Initial Query
Let's first query the LLM directly without any specific data context. A standard question such as “What can H2O help customers with?” will yield a general response based on the vast training data of the LLM.
Using Retrieval-Augmented Generation
Next, we’ll switch to a retrieval-augmented generation approach by asking the same question. The system will now reference the newly embedded data from the H2O website.
By processing the question through the vector database, the platform retrieves the most pertinent information, providing a relevant and detailed answer based on the latest offerings from H2O.
Exploring References and Prompts
You can click on references to see which sections of the website contributed to the answer, allowing for fact-based responses. Additionally, you can examine the LLM prompt that was utilized to generate the answer, noting how specific information was integrated into the reply.
Conclusion
This guide provides a brief overview of how to get started with the H2O GPTe, demonstrating its capabilities in creating data collections, processing various file types, and leveraging both general and context-specific queries. We encourage you to explore the free trial environment, load your own data, and tap into the potential of our large language models.
Keywords
- H2O GPTe
- Enterprise RAG
- Free trial
- Data collection
- Embedding model
- Automatic summarization
- Optical Character Recognition (OCR)
- Retrieval-Augmented Generation (RAG)
FAQ
Q1: How can I access the H2O GPTe free trial?
A1: Visit the H2O website at h2o.ai and log in using your personal Gmail or GitHub account.
Q2: What types of data can I upload to H2O GPTe?
A2: You can upload PDFs, audio files, images, and text files, or you can crawl data directly from a website.
Q3: How does H2O GPTe handle private customer information?
A3: H2O GPTe has guardrails to detect private data, allowing for options to either fail the input or redact sensitive information.
Q4: What is retrieval-augmented generation?
A4: Retrieval-augmented generation (RAG) combines traditional question-answering with specific data retrieval from embedded documents to provide accurate, context-driven answers.
Q5: Can I customize the embedding model used in my data collection?
A5: Yes, you can select from different embedding models to suit your data needs, such as multilingual support.