maestro-project/maestro
An analytical cost model evaluating DNN mappings (dataflows and tiling).
This tool helps hardware architects and deep learning engineers understand the trade-offs of different neural network architectures when deployed on specialized hardware. By inputting details about your deep neural network (DNN) and the processing capabilities of your accelerator, MAESTRO predicts its performance, energy consumption, and memory usage. This allows you to optimize your designs before committing to costly hardware implementations.
247 stars. No commits in the last 6 months.
Use this if you need to analyze and compare how different ways of arranging data (dataflows and tiling) for deep neural networks will perform on a custom hardware accelerator, helping you make informed design decisions.
Not ideal if you are a software developer looking to optimize a pre-built neural network on standard CPUs or GPUs, or if you don't work with custom hardware acceleration for deep learning.
Stars
247
Forks
66
Language
MATLAB
License
MIT
Category
Last pushed
Apr 15, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/maestro-project/maestro"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
fastmachinelearning/hls4ml
Machine learning on FPGAs using HLS
alibaba/TinyNeuralNetwork
TinyNeuralNetwork is an efficient and easy-to-use deep learning model compression framework.
KULeuven-MICAS/zigzag
HW Architecture-Mapping Design Space Exploration Framework for Deep Learning Accelerators
fastmachinelearning/hls4ml-tutorial
Tutorial notebooks for hls4ml
doonny/PipeCNN
An OpenCL-based FPGA Accelerator for Convolutional Neural Networks