Installation

Requirements

All the codes are tested in the following environment:

Python 3.8
PyTorch 2.1.0
CUDA 11.8
spconv v2.x spconv-cu11.8 may contain some errors, change spconv-cu11.6 can work fine.

Install LION codebase

git clone https://github.com/happinesslz/LION.git
pip install -r requirements.txt
python setup.py develop

Install Linear RNN operators

We provide some guidance for some difficult to install linear RNN operators.

Mmaba
- install causal-conv1d by running python -m pip install causal-conv1d==1.2.0.post2
- cd pcdet/ops/mamba
- python setup.py install
RWKV
- Because the hard code of T_MAX and HEAD_SIZE in RWKV, we should change them according to the running config.
- Given the config (maximum group_size: 4096, layer_dim: 128, nhead: 4)
- You should modify T_MAX and HEAD_SIZE in pcdet/ops/wkv6/src/wkv6_cuda.cu and pcdet/models/model_utils/rwkv_cls.py.
- T_MAX = 4096 Because the T_MAX equals with maximum group_size.
- HEAD_SIZE = 128 / 4 = 32 Because the HEAD_SIZE equals layer_dim divided by nhead.
- Rerun python setup.py develop to install the LION codebase.