mmengine-lite

0.10.7 verified Sat May 09 auth: no python

mmengine-lite is a lightweight version of MMEngine, the core engine for OpenMMLab projects. It provides fundamental utilities for deep learning training (config, registry, runner) without heavy dependencies like torch. Current version: 0.10.7. Release cadence: irregular, with minor patches every few months.

pip install mmengine-lite

Common errors

error ModuleNotFoundError: No module named 'mmengine' ↓

cause mmengine-lite is installed but code tries to import from mmengine.

fix

Install mmengine (full version) with pip install mmengine, or change imports to use mmengine-lite compatible paths.

error ImportError: cannot import name 'Config' from 'mmengine' ↓

cause Incorrect import path; Config is now in mmengine.config.

fix

from mmengine.config import Config

Warnings

gotcha mmengine-lite intentionally excludes torch and heavy dependencies. If you need torch (e.g., Runner, optimizers), use the full mmengine package instead (pip install mmengine). ↓

fix Install mmengine instead: pip install mmengine

breaking From mmengine v0.10.0+, the import path for many utilities changed. Previously, some symbols were importable directly from mmengine; now they require submodule imports. ↓

fix Use from mmengine.config import Config instead of from mmengine import Config

gotcha The lite version is not a drop-in replacement for mmengine. Features like Runner, optimizers, and checkpointing require torch and are not available in mmengine-lite. ↓

fix Check the official docs to see which modules are available in the lite version.

Imports

Config
wrong
```
from mmengine import Config
```
correct
```
from mmengine.config import Config
```
mmengine-lite splits modules into separate packages; Config lives in mmengine.config.

Quickstart

Demonstrates using Config to create a config object from a dict.

from mmengine.config import Config

cfg = Config(dict(a=1, b=dict(c=[1,2,3])))
print(cfg.a)