fusionml 0.2.0

High-Performance ML Framework for Apple Silicon

FusionML - Python

High-Performance ML Framework for Apple Silicon with GPU+CPU parallel execution.

Installation

pip install fusionml

# With Metal GPU support
pip install fusionml[metal]

Quick Start

import fusionml as fml

# Initialize
fml.init()

# Create tensors
x = fml.rand(32, 784)
y = fml.Tensor([0, 1, 2, 3])  # Labels

# Build model
model = fml.nn.Sequential([
    fml.nn.Linear(784, 256),
    fml.nn.ReLU(),
    fml.nn.Linear(256, 10)
])

# Optimizer
optimizer = fml.optim.Adam(model.parameters(), lr=0.001)

# Training step
output = model(x)
loss = fml.nn.functional.cross_entropy(output, y)
loss.backward()
optimizer.step()

print(f"Loss: {loss.item()}")

Features

• 🔥 PyTorch-like API
• ⚡ GPU+CPU parallel execution
• 🧠 Full autograd support
• 🎯 Apple Silicon optimized

API

Tensors

fml.rand(2, 3)      # Random uniform
fml.randn(2, 3)     # Random normal
fml.zeros(2, 3)     # Zeros
fml.ones(2, 3)      # Ones
fml.eye(3)          # Identity

Layers

fml.nn.Linear(in, out)
fml.nn.ReLU()
fml.nn.GELU()
fml.nn.Dropout(0.5)
fml.nn.Sequential([...])

Optimizers

fml.optim.SGD(params, lr=0.01, momentum=0.9)
fml.optim.Adam(params, lr=0.001)

Functional

fml.nn.functional.relu(x)
fml.nn.functional.softmax(x)
fml.nn.functional.cross_entropy(pred, target)
fml.nn.functional.mse_loss(pred, target)

License

MIT