aiSSEMBLE Inference YOLO Module

YOLO model family support for the aiSSEMBLE Inference library.

Supported Models

This module supports all YOLO versions available through the Ultralytics library:

Installation

pip install aissemble-inference-yolo

This will also install the required dependencies:

• aissemble-inference-core
• ultralytics (for YOLO models)
• mlserver (for serving)

Usage

With MLServer

Create a model-settings.json:

{
    "name": "yolo",
    "implementation": "aissemble_inference_yolo.YOLORuntime",
    "parameters": {
        "model": "yolov8n.pt"
    }
}

Start MLServer:

mlserver start /path/to/models

With OIP Client

from aissemble_inference_core.client import InferenceClient
from aissemble_inference_core.client.registry import ModuleRegistry

# The module auto-registers via entry points
print(ModuleRegistry.instance().list_available())
# {'runtimes': ['yolo'], 'translators': ['yolo'], ...}

# Use with InferenceClient
client = InferenceClient(adapter, endpoint)
result = client.detect_object("yolo").image("photo.jpg").run()

Components

YOLORuntime

MLServer-compatible runtime that wraps YOLO models:

• Accepts base64-encoded images via OIP protocol
• Returns bounding boxes, labels, and confidence scores
• Configurable model variant via parameters

YOLOTranslator

Translator for YOLO outputs:

• Inherits from DefaultObjectDetectionTranslator
• Can be extended with YOLO-specific optimizations

Entry Points

This module registers the following entry points: