Averion SDK — AI Made Efficient and Open
Complete development environment with optimized libraries, graph compiler, and Python SDK. Get started in 10 minutes with PyTorch or TensorFlow.
10-Minute PyTorch/TensorFlow Onboarding
Deploy your existing models to AliceAI with minimal code changes
Simple Integration
Just add .to('averion') to your existing PyTorch code
import torch
from averion_sdk import AliceAI
# Load your existing model
model = torch.load('my_model.pt')
# Deploy to AliceAI - that's it!
model.to('averion')
# Run inference
output = model(input_data)Four Software Layers
Complete stack from high-level APIs to hardware control
Python SDK
Layer 4
High-level Python interface for rapid prototyping and deployment. Simple model.to('averion') syntax.
- PyTorch integration
- TensorFlow support
- Jupyter notebooks
- Example models
Graph Compiler
Layer 3
Optimizing compiler that transforms ML graphs into efficient AliceAI execution plans.
- Auto-optimization
- Operator fusion
- Memory planning
- Performance profiling
Optimized Kernel Library
Layer 2
Hand-tuned kernels for common AI operations, maximizing hardware utilization.
- GEMM operations
- Convolutions
- Attention layers
- Custom operators
Hardware Abstraction Layer
Layer 1
Low-level interface providing direct access to AliceAI hardware capabilities.
- Memory management
- DMA control
- Interrupt handling
- Power management
Getting Started
Three simple steps to deploy on AliceAI
1
Install Averion SDK
Download and install the SDK package for your development environment.
pip install averion-sdk2
Load Your Model
Import your existing PyTorch or TensorFlow model.
import torch
model = torch.load('my_model.pt')3
Deploy to AliceAI
Simple one-line deployment to AliceAI hardware.
model.to('averion')
output = model(input_data)Framework & Technology Support
Built on open standards and industry-leading frameworks
🔥
PyTorch
🧮
TensorFlow
⚙️
RISC-V
💾
HBM3e
Open Source Commitment
Contributing to TVM and OpenXLA communities for transparent, collaborative development
Apache TVM
Open-source machine learning compiler framework for optimizing models across hardware platforms
OpenXLA
Domain-specific compiler for linear algebra that optimizes TensorFlow and JAX computations
Join Developer Program
Get Early Access
Sign up to receive SDK updates, documentation, and early access to new features