Catwalk

Where AI Models Strut with Style & Performance

The elegant framework for seamless AI model execution across any hardware.
Zero configuration, maximum performance, and intelligent caching.
Let your models walk the runway with confidence.

Get Started View on GitHub

Why Choose Catwalk?

Zero Configuration

One line of code gets you started. Catwalk automatically detects your hardware, selects optimal devices, and applies performance optimizations.

Lightning Fast

Intelligent caching provides 10x faster model loading. ONNX integration delivers 2-5x inference speedup with automatic optimization.

Multi-Format Support

PyTorch, ONNX, HuggingFace, TensorFlow – all through one unified API. Automatic format conversion when beneficial.

Hardware Agnostic

Seamlessly runs on CPU, CUDA, Apple Silicon, and more. Automatic device selection and memory optimization.

Intelligent Caching

Smart model caching with TTL policies, LRU eviction, and integrity checking. Never load the same model twice.

Built-in Benchmarking

Performance profiling, throughput analysis, and optimization insights built right into the framework.

Simple. Elegant. Powerful.

From loading models to running inference, Catwalk makes AI deployment effortless. See how three lines of code deliver enterprise-grade performance.

Try It Now
catwalk_example.py
# Load any model with zero configuration
from pycatwalk import CatwalkRunner

# That's it! Auto-detects format, device, optimizations
runner = CatwalkRunner("model.pt")
results = runner.predict(your_data)

# Works with PyTorch, ONNX, HuggingFace, TensorFlow onnx_runner = CatwalkRunner("model.onnx")
hf_runner = CatwalkRunner("bert-base-uncased")

# Advanced configuration when you need it
from pycatwalk import ModelConfig
config = ModelConfig(
    use_mixed_precision=True,
    enable_compilation=True,
    convert_to_onnx=True, # Auto-convert for speed
    cache_model=True # 10x faster reloads
)
runner = CatwalkRunner("model.pt", config=config)

Performance That Matters

10x
Faster Model Loading
5x
ONNX Inference Speedup
34%
Memory Reduction
98.5%
Optimal Device Selection

Get Started in Minutes

1

Install Catwalk

pip install pycatwalk

Optional accelerators:
pip install onnxruntime-gpu
for ONNX GPU support
pip install transformers
for HuggingFace models

2

Load Your Model

Point Catwalk to any model file or HuggingFace model name. It automatically detects the format and applies optimal settings for your hardware.

3

Run Inference

Use the simple predict() method with any input format. Catwalk handles batching, device placement, and performance optimization automatically.