Installation Guide

SAGA is designed to be lightweight, highly optimized, and easy to integrate into any existing PyTorch-based deep vision pipeline.

Prerequisites

Python: 3.10 or newer
PyTorch: 2.0.0 or newer (CUDA support highly recommended)
Triton: 2.1.0 or newer (Optional, Linux/NVIDIA only. Required for fused kernel acceleration)

Installing via PyPI

The easiest way to install SAGA is directly from the Python Package Index.

For standard CPU/GPU usage (Eager Mode):

pip install saga-activation

For High-Speed GPU Execution (Linux & NVIDIA GPUs Only):

To unlock the fused memory-bandwidth optimizations, install SAGA with the Triton extension:

pip install "saga-activation[triton]"

Installing from Source

If you are working in a standard local environment, clone the repository and install it in editable mode:

git clone [https://github.com/sijuswamyresearch/saga-activation.git](https://github.com/sijuswamyresearch/saga-activation.git)
cd saga-activation

# Standard installation
pip install -e .

# Installation with Triton acceleration and development tools
pip install -e ".[dev,triton]"

If you are testing SAGA in a notebook environment, you must use the shell prefix (!) and directory magic (%) to install the package directly within a cell:

!git clone [https://github.com/sijuswamyresearch/saga-activation.git](https://github.com/sijuswamyresearch/saga-activation.git)
%cd saga-activation
!pip install -e ".[triton]"

Verifying the Installation

To verify that SAGA is installed and your GPU is picking it up correctly, run the following diagnostic script. This will also confirm if the new v0.2.0 interpretability mode and Triton fallback are working.

import torch
from saga import SAGA

# Device-agnostic setup
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f"Running on: {device}")

# Initialize SAGA with interpretability (return_gate=True)
act = SAGA(in_channels=64, return_gate=True).to(device)
x = torch.randn(1, 64, 128, 128).to(device)

# Pass the tensor through the activation
out, gate = act(x)

print(f"Output shape: {out.shape}")      # Should be: torch.Size([1, 64, 128, 128])
print(f"Gate map shape: {gate.shape}")   # Should be: torch.Size([1, 64, 128, 128])