Selective-SSM

Experimental exploration of applying Selective State Space Model (S-SSM) architectures within Amortized Bayesian Inference workflows

Requirements:

• Linux / WSL
• Python 3.11+
• PyTorch 1.12+
• NVIDIA GPU
• CUDA 11.6+

Installation

First create a new conda environment with at least Python 3.11 support

conda create -n bf-ssm python=3.11

Install libraries (should use .yaml env for this)

conda install numpy pandas matplotlib seaborn ipykernel

The conda forge index is currently behind, so we'll have to use pip for the more prominent libraries

pip install torch
pip install keras
pip install triton
pip install mamba-ssm

Install development build of BayesFlow

pip install git+https://github.com/Chase-Grajeda/BayesFlow@ssm-wrapper