BigDL Nano

2.4.0 verified Fri May 01 auth: no python deprecated

BigDL Nano is a component of the BigDL library that provides high-performance, scalable acceleration for Intel architectures, including optimizations for PyTorch and TensorFlow model training and inference. Current version is 2.4.0. Release cadence has been irregular, with major releases ~annually, but recently shifted to the IPEX-LLM project.

pip install bigdl-nano

Common errors

error ModuleNotFoundError: No module named 'bigdl_nano' ↓

cause Incorrect import path; using underscore instead of dot.

fix

Use: from bigdl.nano import Nano

error ImportError: cannot import name 'Nano' from 'bigdl.nano' ↓

cause The package is not installed correctly or version mismatch.

fix

Install the latest version: pip install --upgrade bigdl-nano

Warnings

deprecated BigDL Nano is deprecated in favor of IPEX-LLM (Intel Extension for PyTorch for Large Language Models). New projects should use IPEX-LLM. ↓

fix Switch to ipex-llm: pip install ipex-llm and use from ipex_llm import ...

breaking BigDL Nano 2.4.0 drops support for Python 3.7 and earlier. ↓

fix Upgrade to Python 3.8+.

Imports

Nano

wrong

from bigdl_nano import Nano

correct

from bigdl.nano import Nano

Use dots, not underscores, for import paths.

Quickstart

Basic usage: import Nano, create an instance, and optimize a PyTorch model.

from bigdl.nano import Nano
import torch
nano = Nano()
model = torch.nn.Linear(10, 2)
optimized_model = nano.optimize(model)