Inference Models

0.27.2 verified Fri May 01 auth: no python

The new inference engine for Computer Vision models, providing fast, optimized inference for object detection, classification, and segmentation. Current version 0.27.2, under active development.

pip install inference-models

Common errors

error ModuleNotFoundError: No module named 'inference' ↓

cause The top-level module was renamed from 'inference' to 'inference_models' in version 0.20.0.

fix

Run 'pip install inference-models' and use 'from inference_models import ...'.

error ImportError: cannot import name 'InferencePipeline' from 'inference_models' ↓

cause Incorrect import path or outdated version. InferencePipeline is available from version 0.20.0+.

fix

Upgrade to latest: 'pip install --upgrade inference-models' and use 'from inference_models import InferencePipeline'.

error TypeError: load_model() missing 1 required positional argument: 'model_id' ↓

cause The 'load_model' function requires a model identifier (e.g., Hugging Face ID or path).

fix

Provide a model_id: model = load_model('facebook/detr-resnet-50')

Warnings

breaking In version 0.20.0, the 'inference' top-level module was renamed to 'inference_models'. Old imports like 'import inference' will fail. ↓

fix Use 'from inference_models import ...' instead.

deprecated The function 'run_inference()' is deprecated since 0.25.0 and will be removed in 0.30.0. Use 'InferencePipeline' instead. ↓

fix Replace with: pipeline = InferencePipeline(model); results = pipeline(image)

gotcha The library requires Python >=3.10 and <3.13. It does not support Python 3.9 or earlier, and will not install on Python 3.13+. ↓

fix Ensure your Python version is 3.10, 3.11, or 3.12.

Imports

InferencePipeline

wrong

from inference_models.pipeline import InferencePipeline

correct

from inference_models import InferencePipeline

InferencePipeline is exposed at package level; submodule import may break.

load_model

wrong

import inference_models.load_model

correct

from inference_models import load_model

load_model is a function, not a module.

Quickstart

Load a model and run inference on an image.

import os
from inference_models import InferencePipeline, load_model

# Replace with your model path or Hugging Face ID
model = load_model('facebook/detr-resnet-50')

# Create pipeline
pipeline = InferencePipeline(model)

# Run inference
image_path = 'path/to/image.jpg'
results = pipeline(image_path)
print(results)