Welcome to Gati Platform DocumentationΒΆ
Gati is a complete FPGA-based deep learning inference ecosystem designed to accelerate Convolutional Neural Networks (CNNs) on the Vaaman platform. It combines a high-performance hardware accelerator (GATI) with a software toolchain (GATICC) that enables users to compile, deploy, simulate, and execute machine-learning models.
The project is designed around a simple workflow:
Train or obtain a machine-learning model in ONNX format.
Use GATICC to compile and optimize the model.
Generate a hardware configuration for the target model.
Program the FPGA with the appropriate GATI bitstream.
Run accelerated inference on the Vaaman SBC.
By combining FPGA acceleration with a flexible software stack, Gati provides a platform for deploying low-latency and power-efficient neural network inference workloads. The system supports multiple hardware architectures and a growing set of neural network operators, allowing users to execute a wide range of CNN-based models on FPGA hardware.
The ecosystem consists of two major components:
GATI The FPGA hardware accelerator responsible for executing neural network operations. GATI implements the compute engines, memory architecture, and data movement required to perform accelerated CNN inference on the FPGA.
GATICC The accompanying software toolchain that compiles ONNX models, manages deployment, provides simulation capabilities, and exposes a Python API for interacting with the accelerator.
Together, GATI and GATICC provide an end-to-end workflow that takes a machine-learning model from development to FPGA deployment with minimal user effort.
Whether you are evaluating existing neural networks, developing custom CNN architectures, or exploring FPGA-based machine learning acceleration, the Gati ecosystem provides the tools necessary to move from ONNX models to accelerated hardware execution.