neuratensor 1.0.0

Production neuromorphic inference SDK with CUDA acceleration

NeuraTensor SDK

Production neuromorphic inference at 8ms latency

NeuraTensor is a high-performance neuromorphic inference SDK featuring a proprietary fused SNN-SSM architecture with CUDA acceleration.

⚡ Performance

• 8.13ms latency (64M model, Jetson AGX Orin)
• 123 seq/s throughput
• 120x faster than PyTorch baseline
• Sub-10ms real-time inference

�� Quick Start

pip install neuratensor

from neuratensor import NeuraTensor, NeuraTensorConfig

# Load model
config = NeuraTensorConfig.preset("64m")
model = NeuraTensor(config).cuda().half()

# Inference
output = model(input_ids)
# ✅ 8ms latency, 64M params, 123 seq/s

📊 Models

Model	Parameters	Latency	Use Case
64M	64M	8ms	Edge, real-time
256M	256M	~15ms	Balanced
1B	1B	~35ms	Quality

🔧 Requirements

• NVIDIA GPU (CUDA 11.4+, SM 7.0+)
• Python 3.8+
• PyTorch 2.0+

📚 Documentation

• Distribution Guide
• Patent Notice

📄 License

Proprietary. Binary distribution only.

© 2024-2025 Neuramorphic, Inc.