Built on multi-modal data fusion, the Lindorm AI engine provides a one-stop service for data storage, computing, and model inference. This allows you to perform AI inference tasks, such as semantic search, multi-modal retrieval, and knowledge-based Q&A, directly within the database without exporting data to other AI platforms. This ensures that searches and Q&A are always based on the latest data while meeting security, privacy, and regulatory requirements.
With the Lindorm AI engine, you can perform the following tasks:
Vectorization for text or images Convert text or images in the wide table engine into vectors for scenarios such as vector search and clustering.  |
Semantic similarity calculation and reranking Calculate the semantic similarity between a target text and the search results from Lindorm Search, and then rerank the results from high to low by similarity score.  |
The service finds information semantically similar to a user's question in Lindorm Search and the wide table engine. After reranking the search results, it combines them with the user's question to form a prompt sent to a large model. This creates a private knowledge-based Q&A service built on your enterprise data.  |
Get started with the Lindorm AI engine
To quickly try out data vectorization, semantic similarity calculation, reranking, and private knowledge-based Q&A, see Quickly build an intelligent search service based on the multi-modal capabilities of Lindorm.
Why choose the Lindorm AI engine
Keep data in-database for one-stop data processing and AI inference
You can perform all tasks, from data ingestion and storage to computing and inference, within Lindorm. This keeps your data within the database and helps you meet security, privacy, and regulatory requirements.
You do not need to build data pipelines from Lindorm to other AI platforms. This reduces system complexity and maintenance costs. In addition, you can run inference tasks directly on the latest data in Lindorm, which improves data freshness and the accuracy of inference results.
Elastic heterogeneous computing to improve AI inference performance
The inference nodes of the Lindorm AI engine support multiple instance types (CPU and GPU). You can use CPUs for traditional inference tasks and leverage GPUs to accelerate complex inference workloads, enhancing overall performance. In addition, inference nodes share storage with multi-modal engines. This architecture reduces data transfer overhead, enables near-data inference optimization, and improves inference efficiency.
Seamless integration with external model platforms and ecosystems
The engine supports one-click import of models from ModelScope and Hugging Face, simplifying model integration and allowing you to focus on application development and model optimization. You can also upload your own custom models to meet specific business needs.
Simplified interaction with native database SQL
With Lindorm SQL, you can create AI models and run end-to-end inference without mastering advanced programming languages.
Supported models
The Lindorm AI engine supports deploying open-source models from ModelScope and Hugging Face. You can also upload a custom model.
Model type | Model list | Model platform |
Vectorization (Embedding) |
| Hugging Face |
jina-embeddings-v2-base-zh | ModelScope | |
Reranking (ReRank) |
| Hugging Face |
Large language model (LLM) | ChatGLM-6B | Hugging Face |
| ModelScope | |
Text-to-image generation | Stable Diffusion | ModelScope |
Custom model | N/A | N/A |
Billing
To use the Lindorm AI engine, you need AI foundation nodes or AI engine nodes. AI engine nodes are a paid service, whereas AI foundation nodes and model calls are free of charge.
The following table describes the required nodes and billing details for different use cases.
Use case | Required nodes | Billing |
AI inference services, including feature vector extraction, semantic search and reranking, Q&A, text generation, and text-to-image | AI foundation node | Free |
AI engine node | Paid. For pricing details, see the purchase page. |
AI foundation nodes manage and coordinate the AI engine. AI engine nodes deploy inference services.