snn-spike 1.0.0

Spike: Super Neural Network C++ Engine with 0.25-bit AVX2 Quantization

Spike

Super Neural Network (SNN) Core Engine with 0.25-bit AVX2 Quantization

---

Spike is a high-performance Spiking Neural Network (SNN) core and language modeling framework designed for extreme efficiency. It combines a hardware-accelerated C++ core optimized with AVX2 instruction sets and a biologically-inspired Python layer that models parameters as living, metabolic creatures with lateral inhibition and dynamic rebirth cycles.

Features

Subsystem	Description
0.25-bit Quantization	Sub-bit Bloom-filter based weight quantization for ultra-low memory footprints.
Metabolic Parameters	Parameters behave like living cells with energy consumption, regeneration, and hunger.
Holographic LM Head	Bypasses standard projection layers via holographic token mapping.
Associative Scans	Lightning-fast parallel recurrent associative state scans for sequence modeling.
IPP Rebirth Protocol	Intelligence Per Parameter tracking that recycles dead parameters during training.

Installation

You can install Spike directly from PyPI:

pip install spike

From source (requires a C++ compiler supporting C++11/C++14/C++17):

git clone https://github.com/mathagens-ai/spike.git
cd spike
pip install -e .

Quick Start

import spike
import numpy as np

# Instantiate standard SNN configuration
config = spike.SNNConfig.SNN_Nano()
print(f"Loaded config: vocab_size={config.vocab_size}, d_model={config.d_model}")

# Instantiate model
model = spike.SNNModel(config)

# Compile model weights down to 0.25-bit/0.45-bit Bloom Tensors
dummy_weights = {
    "encoder_attn": np.random.randn(256, 256).astype(np.float32),
    "ffn": np.random.randn(256, 512).astype(np.float32)
}
model.compile_from_dense(dummy_weights)

# Run a forward pass
input_ids = np.random.randint(0, config.vocab_size, (1, 32))
logits, states = model.forward(input_ids)

print("Forward pass successful. Logits shape:", logits.shape)

Running Tests

To verify the installation:

pytest tests/

License

Apache License 2.0 — see LICENSE for details.

Copyright 2026 Mathagens AI