AutoDL framework for neural network compression & acceleration
up to
25 times
neural network compression
2 weeks
up to
70%
hardware cost
reduction
neural network acceleration
20 times
up to
to a compressed
model
Product
ENOT is a framework with a Python API that can be quickly and easily integrated within various neural network training pipelines
Language
CNN
RNN
LSTM
DNN
Machine Learning Frameworks
Deployment
NN Types
Hardware Libraries
CPU
FGPA
GPU
NPU
Runtime
On-premises
On the cloud
Cases
Industry
Processing images faster
Client required faster image processing on smartphones for better user experience without sacrificing accuracy.
Case
Smartphone manufacturer
4.8 times
acceleration without accuracy degradation
Reduction of cloud costs
Client was facing high cloud server costs for their facial recognition pipeline, thus their neural networks were accelerated.
Case
Results
3.2 times
reduction in cloud infrastructure costs
Low latency & RAM consumption
Case
Client required to reduce their NN model size to meet their RAM limitations, while maintaining low latency
Results
4.8 Mb
NN model size
9.1Mb
Peak RAM consumption
4 ms
Latency
Meet chipset's RAM limitations
Client required compression of their neural network to meet their chipset's 5MB RAM limitation
Case
Results
Reduction of hardware costs
Case
Client was incurring very high hardware costs from operating object detection on 25 video streams.
Results
4.2 times
Acceleration
2.2 times
Reduction in server costs
8 times
Reduction of peak
RAM consumption
from 36 Mb to 4.5 Mb
Faster facial keypoint detection
Client could not achieve fast enough facial keypoint detection to provide a seamless mobile app experience.
Case
Results
48%
Acceleration of the baseline model on multiple mobile platforms
Results
Telecommunications
Smartphone manufacturer
Electronics manufacturer
AI-based mobile app
Oil & Gas
Electronics
Healthcare
Oil & Gas
Autonomous Driving
Cloud Computing
Telecom
Mobile Apps
Internet of Things
Robotics
Technology
Our neural network architecture search engine allows to automatically find the best possible architecture from millions of available options, taking into account several parameters:
— input resolution
— depth of neural network
— operation type
— activation type
— number of neurons at each layer
— bit width for target hardware platform for NN inference
NAS
Pruning
Quantization
Distillation
02
01
03
04
ENOT applies these methods simultaneously to achieve the highest compression/acceleration rate without accuracy degradation. It allows to automate the search for the optimal neural network architecture, taking into account latency, RAM and model size constraints for different hardware and software platforms.
Try ENOT
Pricing
Don't compromize on features.
Process locally, deploy anywhere, optimise as many models as you want.
Simultaneous vRAM usage during optimization
<25GB
Number of optimized models
unlimited
Target hardware
any hardware (CPU/GPU)
Deployment on cloud/premises
Data privacy and protection
local storage and processing
available
60$/month
Simultaneous vRAM usage during optimization
<100GB
Number of optimized models
unlimited
Target hardware
any hardware (CPU/GPU)
Deployment on cloud/premises
Data privacy and protection
local storage and processing
available
55$/month
Simultaneous vRAM usage during optimization
>100GB
Number of optimized models
unlimited
Target hardware
any hardware (CPU/GPU)
Deployment on cloud/premises
Data privacy and protection
local storage and processing
available
45$/month
Per GB of vRAM
Per GB of vRAM
Per GB of vRAM
Pricing
Don't compromise on features.
Process locally, deploy anywhere, optimize as many models as you want.
Price per GB of vRAM
60$/month
55$/month
45$/month
Simultaneous vRAM usage during optimization
Number of optimization models
Target hardware
Deployment on Cloud
Deployment on Premises
Inference
<25GB
<100GB
>100GB
unlimited
unlimited
unlimited
any hardware (CPU/GPU)
any hardware (CPU/GPU)
any hardware (CPU/GPU)
local storage and processing
local storage and processing
local storage and processing
the monthly license plan is tied to your machine's total GPU vRAM usage during training.
*
*
even if you stop your subscription,
optimized models are free to use forever
Data privacy and protection
popular choice
Our Partners
Chief Executive Officer
Serge Aliamkin
Our Team
Chief technology Officer
Alex Goncharenko
Chief Financial Officer
Michael Berkov
Strategic Partnerships
Vlad Dyubanov
Business Development
Kristians Karlsons
Chief Research Officer
Ivan Oseledets
Close
Do you have any questions? Contact us!