sintellix 0.1.1

High-Performance Neural Network Framework with 3D Grid Architecture

Sintellix Python API

PyTorch-like Python interface for Sintellix Neural Network Framework with HOT (Higher Order Thought) Architecture.

Features

• PyTorch-like API: Familiar interface for PyTorch users
• CUDA Acceleration: High-performance CUDA kernels for all operations
• Advanced Architecture: Multi-head attention, SSM, RWKV, temporal attention, DDPM, and more
• Tiered Storage: Automatic GPU→RAM→Disk memory management
• VQ-GAN Codec: Semantic encoding/decoding with vector quantization
• Easy Training: Built-in trainer with checkpointing and logging

Installation

pip install sintellix

Or install from source:

git clone https://github.com/sintellix/sintellix.git
cd sintellix/python
pip install -e .

Quick Start

Basic Usage

from sintellix import NeuronModel, NeuronConfig

# Create model
config = NeuronConfig(dim=256, grid_size=(32, 32, 32))
model = NeuronModel(config)
model.initialize()

# Forward pass
import torch
input_tensor = torch.randn(1, 256, 256)
output = model(input_tensor)

Training

from sintellix import Trainer, TrainingConfig

# Create trainer
train_config = TrainingConfig(
    learning_rate=0.001,
    batch_size=32,
    epochs=100
)
trainer = Trainer(model, train_config)

# Train
trainer.train(train_loader, val_loader)

Model Management

from sintellix import download_model

# Download pretrained models
e5_path = download_model("e5-large")
vqgan_path = download_model("vqgan-codebook")

Configuration

config = NeuronConfig(
    dim=256,                    # Neuron dimension
    num_heads=8,                # Attention heads
    grid_size=(32, 32, 32),     # Neuron grid
    temporal_frames=8,          # Temporal history
    enable_multi_head=True,     # Enable modules
    enable_ssm=True,
    enable_rwkv=True,
    gpu_cache_size_mb=4096,     # Storage config
    ram_cache_size_mb=16384,
)

License

MIT License - see LICENSE for details