API Overview

​

Overview

​

Main Modules

​

Quick Start

import hls4ml

# Create configuration
config = hls4ml.utils.config.create_config(
    output_dir='my-hls-test',
    project_name='myproject',
    backend='Vivado'
)

# Convert Keras model
hls_model = hls4ml.converters.convert_from_keras_model(
    model,
    output_dir='my-hls-test',
    project_name='myproject',
    backend='Vivado'
)

# Compile and predict
hls_model.compile()
predictions = hls_model.predict(test_data)

​

Core Functions

​

Model Conversion

• convert_from_keras_model() - Convert Keras models
• convert_from_pytorch_model() - Convert PyTorch models
• convert_from_onnx_model() - Convert ONNX models
• convert_from_config() - Convert from YAML configuration

​

Configuration

• create_config() - Create backend configuration
• config_from_keras_model() - Generate config from Keras model
• config_from_pytorch_model() - Generate config from PyTorch model
• config_from_onnx_model() - Generate config from ONNX model

​

Model Operations

• model.compile() - Compile the generated project
• model.predict() - Run inference
• model.build() - Synthesize with HLS tools
• model.write() - Write project to disk

​

Module Structure

hls4ml/
├── converters/          # Model conversion modules
│   ├── keras/          # Keras converters
│   ├── pytorch/        # PyTorch converters
│   └── onnx/           # ONNX converters
├── model/              # Model graph representation
│   ├── graph.py        # ModelGraph class
│   └── optimizer/      # Optimization passes
├── backends/           # FPGA backends
│   ├── vivado/         # Xilinx Vivado backend
│   ├── vitis/          # Xilinx Vitis backend
│   └── quartus/        # Intel Quartus backend
└── utils/              # Utilities
    ├── config.py       # Configuration helpers
    └── report.py       # Report parsing

​

Architecture

​

Conversion Flow

1. Parse - Extract model architecture and weights
2. Transform - Convert to internal representation (ModelGraph)
3. Optimize - Apply optimization passes
4. Generate - Produce HLS C++ code
5. Build - Synthesize with backend tools

​

ModelGraph

• Stores layers as nodes in a directed graph
• Manages data flow between layers
• Applies optimization transformations
• Generates backend-specific code

​

Type System

• FixedPrecisionType - Fixed-point types (e.g., ap_fixed<16,6>)
• IntegerPrecisionType - Integer types (e.g., ap_int<8>)
• FloatPrecisionType - Floating-point types
• UnspecifiedPrecisionType - Auto-inferred precision

​

Next Steps