Skip to content

Latest commit

 

History

History
163 lines (121 loc) · 5.95 KB

readme.md

File metadata and controls

163 lines (121 loc) · 5.95 KB
Raw

Example Code for REST API Calls and Vector Calculations in SQL Server

This repository contains example code for demonstrating REST API integration and vector manipulation within a SQL Server context. Note: This is development-level code and is not intended for production use. It is meant to showcase how to implement these features in SQL Server using SQL CLR assemblies.

Table of Contents

What's new

  • Added method to calculate vector size.
  • Added function for local embeddings calculation using Ollama models.
  • Added sample for Azure OpenAI API integration.

Features

  1. REST API Integration:

    • myRestEndpoint.cs: Provides a SQL CLR stored procedure to invoke REST APIs from SQL Server. Supports:
      • GET and POST methods.
      • Custom headers in JSON format.
      • Payload handling for POST requests.

  2. Vector Type and Operations:

    • myVector.cs: Implements a SQL CLR User-Defined Type (UDT) for vectors, allowing:
      • Serialization and deserialization of vectors.
      • Support for multiple distance metrics (Cosine, Euclidean, Manhattan, and (dot) Negative Inner Product).
      • Usage within SQL Server for advanced calculations.

Files in this Repository

REST API Integration

  • myRestEndpoint.cs:

    • SQL CLR stored procedure for invoking REST APIs.
    • Handles JSON-formatted headers and POST payloads.
    • Supports TLS 1.2 for secure communication.

  • myRestEndpoint.csproj:

    • Project file for compiling the REST API integration code.

Vector Operations

  • myVector.cs:

    • SQL CLR UDT for vector operations.
    • Supports parsing from JSON-like strings and performing vector distance calculations.
    • Implements serialization for SQL Server compatibility.
    • Assesmbly deployed in SAFE mode.

  • myVector.csproj:

    • Project file for compiling the vector operations code.
    • Assesmbly deployed in EXTERNAL_ACCESS mode.

  • myVector.sln:

    • Solution file to manage both the REST API and vector projects in Visual Studio.

Requirements

  • Visual Studio 2022 (or newer).
  • .NET Framework 4.8.
  • SQL Server edittion with CLR support.
  • The UDF needs EXTERNAL_ACCESS. SQL CLR in linux supports only SAFE assemblies link.

Limitations

  • Vector type:
    • Cannot define vector size during table creation.
    • Requires dbo. prefix when invoking vector functions.
  • Performance of REST API and vector operations in production scenarios has not been analyzed.
  • REST API integration disables certificate validation in testing for convenience (should not be used in production).

Compile

  • dotnet build -c Release myRestEndpoint.csproj
  • dotnet build -c Release myVector.csproj

Usage

  1. REST API Calls: Deploy the myRestEndpoint assembly to your SQL Server and use the sp_invoke_external_rest_endpoint2 stored procedure to call REST APIs directly from SQL Server.

    Example for Chat Completion:

DECLARE @response NVARCHAR(MAX);
DECLARE @chat_completion NVARCHAR(MAX);

EXEC dbo.sp_invoke_external_rest_endpoint2
    @url = 'https://xxxxx.openai.azure.com/openai/deployments/gpt-4o-mini/chat/completions?api-version=2023-05-15',
    @method = 'POST',
    @payload = N'{"messages": [{"role": "system", "content": "Your system message"}, {"role": "user", "content": "Your user prompt"}], "temperature": 0.1, "max_tokens": 1000}',
    @headers = '{"Content-Type":"application/json", "api-key":"zzzzzz"}',
    @response = @response OUTPUT;

SET @chat_completion = CAST(JSON_VALUE(@response, '$.choices[0].message.content') AS NVARCHAR(MAX));
SELECT @chat_completion AS Content;

Example for Azure OpenAI Embeddings:

DECLARE @response NVARCHAR(MAX);
DECLARE @vector NVARCHAR(MAX);

EXEC dbo.sp_invoke_external_rest_endpoint2
    @url = 'https://xxxxxx.openai.azure.com/openai/deployments/gpt-4o-mini/embeddings?api-version=2023-05-15',
    @method = 'POST',
    @payload = N'{"input": "Your input text"}',
    @headers = '{"Content-Type":"application/json", "api-key":"zzzzzz"}',
    @response = @response OUTPUT;

WITH ParsedEmbedding AS (
    SELECT value
    FROM OPENJSON(@response, '$.data[0].embedding')
)
SELECT @vector = STRING_AGG(value, ',')
FROM ParsedEmbedding;

SELECT @vector AS Embedding;

  1. Local Embeddings Calculation: Use the sp_invoke_ollama_model stored procedure to calculate embeddings locally with a compatible endpoint.

    Example for Local Embeddings:

DECLARE @response NVARCHAR(MAX);
DECLARE @vector VECTOR;

EXEC dbo.sp_invoke_ollama_model 
    @endpoint = 'http://localhost:11434/api/embeddings',
    @model = 'all-minilm',
    @prompt = 'The sky is blue because of Rayleigh scattering',
    @response = @response OUTPUT;

WITH ParsedEmbedding AS (
    SELECT value
    FROM OPENJSON(@response, '$.embedding')
)
SELECT @vector = CAST(STRING_AGG(value, ',') AS VECTOR)
FROM ParsedEmbedding;

SELECT 
    @vector AS value, 
    @vector.Size() AS size, 
    CAST(@vector AS NVARCHAR(MAX)) AS text;
  1. Vector Calculations:
    • Deploy the Vector UDT to your SQL Server.
    • Use the VECTOR_DISTANCE function to calculate distances between vectors.
    • Example SQL query:
DECLARE @v1 Vector = '[1.0, 2.0, 3.0]';
DECLARE @v2 Vector = '[4.0, 5.0, 6.0]';
SELECT dbo.VectorDistance('euclidean', @v1, @v2) AS Distance;

Pending Work

  • Optimize performance for large-scale use cases.

Disclaimer

This repository is for educational and development purposes only. It is not designed for use in production environments. Use it at your own risk.

License

This project is licensed under the MIT License. See the LICENSE file for details.