Haystack-OPEA

This package contains the Haystack integrations for OPEA Compatible OPEA Microservices. At its core, OPEA offers a suite of containerized microservices—including components for LLMs, embedding, retrieval, and reranking—that can be orchestrated to build sophisticated AI workflows like Retrieval-Augmented Generation (RAG). These microservices are designed for flexibility, supporting deployment across various environments such as cloud platforms, data centers, and edge devices.​

For more information, see Getting Started with OPEA, OPEA Components and Full Examples.

Installation

You can install Haystack OPEA package in several ways:

Install from Source

To install the package from the source, run:

pip install poetry && poetry install --with test

Install from Wheel Package

To install the package from a pre-built wheel, run:

1. Build the Wheels: Ensure the wheels are built using Poetry.

   poetry build

2. Install via Wheel File: Install the package using the generated wheel file.

   pip install dist/haystack_opea-0.1.0-py3-none-any.whl

Examples

See the Examples folder; it contains two jupyter notebooks, using an OPEA LLM and text embedder. The folder also includes a docker compose configuration for starting the OPEA backend.

Embeddings

The classes OPEADocumentEmbedder and OPEATextEmbedder are introduced.

from haystack_opea import OPEATextEmbedder

text_to_embed = "I love pizza!"

text_embedder = OPEATextEmbedder(api_url="http://localhost:6006")
text_embedder.warm_up()

print(text_embedder.run(text_to_embed)

And similarly:

from haystack import Document
from haystack_opea import OPEADocumentEmbedder

doc = Document(content="I love pizza!")

document_embedder = OPEADocumentEmbedder(api_url="http://localhost:6006")
document_embedder.warm_up()

result = document_embedder.run([doc])
print(result["documents"][0].embedding)

LLMs

The class OPEAGenerator is introduced:

from haystack_opea import OPEAGenerator

generator = OPEAGenerator(
    "http://localhost:9009",
    model_arguments={
        "temperature": 0.2,
        "top_p": 0.7,
        "max_tokens": 1024,
    },
)
generator.warm_up()
result = generator.run(prompt="What is the answer?")

For more information, see Haystack Docs and OPEA.