Symfony Semantic Search Demo Application

The "Symfony Semantic Search Demo Application" is a reference application created to show how to develop semantic search following the Symfony Best Practices based on the Symfony Demo application.

You can also learn about these practices in the official Symfony Book.

Requirements

• PHP 8.2.0 or higher;
• PDO-SQLite PHP extension enabled;
• and the usual Symfony application requirements.
• Ollama (along with bge-large:335m-en-v1.5-fp16, all-minilm:22m, snowflake-arctic-embed:335m and llama3.2:1b-instruct-q8_0 models)
• Docker

Usage

Launching Meilisearch and PHP containers:

make up

Serving the application (without Docker):

symfony serve OR php -S localhost:8000 -t public/

Then access the application in your browser at the given URL (https://localhost:80 by default).

Makefile

The make recipes can be accessed using:

make

Meilisearch

The meilisearch service is available at <http://localhost:7700>.

A dump is provided to test the query easily without rebuilding the indexes.

The following commands can be used to manage the meilisearch service:

make enable-vector-store

If you prefer to test it, the following query can be sent (after pulling the models using Ollama) to trigger the embedding generation:

PATCH http://localhost:7700/indexes/app_dev_posts/settings
Content-Type: application/json
Authorization: Bearer 7d9b594befb76b801dd850fd21bc9409174cfc2af41ca3ceda5681ba81f9

{
    "embedders": {
      "ollama_bge-large": {
        "source": "ollama",
        "url": "http://host.docker.internal:11434/api/embeddings",
        "model": "bge-large:335m-en-v1.5-fp16",
        "dimensions": 1024,
        "documentTemplate": "A movie titled {{doc.title}} whose resumed in {{doc.summary}} and the whole critic is {{doc.content|truncatewords: 100}}, the critic is written by {{doc.author.fullName}}",
        "distribution": {
          "mean": 0.9,
          "sigma": 0.4
        }
      },
      "ollama_mini-lm": {
        "source": "ollama",
        "url": "http://host.docker.internal:11434/api/embeddings",
        "model": "all-minilm:22m",
        "dimensions": 384,
        "documentTemplate": "A movie titled {{doc.title}} whose resumed in {{doc.summary}} and the whole critic is {{doc.content|truncatewords: 100}}, the critic is written by {{doc.author.fullName}}",
        "distribution": {
          "mean": 0.75,
          "sigma": 0.25
        }
      },
      "ollama_snowflake": {
        "source": "ollama",
        "url": "http://host.docker.internal:11434/api/embeddings",
        "model": "snowflake-arctic-embed:335m",
        "dimensions": 1024,
        "documentTemplate": "A movie titled {{doc.title}} whose resumed in {{doc.summary}} and the whole critic is {{doc.content|truncatewords: 100}}, the critic is written by {{doc.author.fullName}}",
        "distribution": {
          "mean": 0.75,
          "sigma": 0.25
        }
      },
      "ollama_llama32": {
        "source": "ollama",
        "url": "http://host.docker.internal:11434/api/embeddings",
        "model": "llama3.2:1b-instruct-q8_0",
        "dimensions": 2048,
        "documentTemplate": "A movie titled {{doc.title}} whose resumed in {{doc.summary}} and the whole critic is {{doc.content|truncatewords: 100}}, the critic is written by {{doc.author.fullName}}",
        "distribution": {
          "mean": 0.9,
          "sigma": 0.5
        }
      }
    }
}

PS: The model used can be changed in BlogSearchComponent::getPosts() method. PS II: For more information about "vector search" in Meilisearch, head to the official documentation.