MIL
ABOUT
Research
Product
Divisions
Research
Engineering
MVP
Science Club
Consulting
BLOG
RU/EN
COMPRESSA:
DL Models Compression Platform
Production-ready solution for infrastructure optimization
in several weeks
, instead of several months
Great variability
of supported model architectures
ML Infrastructure
cost reduction
by decreasing the computational complexity of DL models
On-device
ML Infrastructure transfer by compression the DL models into device limitations
BUSINESS CASES
1
RAM, Energy, CPU/GPU lower consumption while models inference
2
Transfer high-quality and complex models on device
3
Models inference on low-bit CPU
4
Speeding up calculations
Learn More about Platform
SOLVING METHODS
Post-training and Low-Bit Quantisation (Adaround, GDRQ, LSQ, own modification of LSQ, APoT, Symmetric, Asymmetric, etc)
Pruning and Knowledge Distillation (HRank, CUP, Cluster-based, Magnitude)
Device Placement
DL Optimization Solvers
EXPERIENCE & EXAMPLES
10-time decreasing of GPU resources while ranking architecture due to the quality in evolutionary NAS methods
Clients Cases
MIL Team Research - Compression Team
Filter level pruning
Seminar
MIL Team Research - Compression Team
Denoising
Models transfer on software chips for wearable devices and smartphones
Clients Cases
MIL Team Research - Compression Team
Quantization experiments
Seminar
MIL Team Research - Compression Team
PyNet, HRNet architectures pruning
Clients Cases
MIL Team Research - Compression Team
Models optimization techniques
Seminar
MIL Team Research - Compression Team
Want your Platform
Заказать проект
Для начала проекта нам нужно поговорить. Достаточно заполнить поля ниже, и мы свяжемся с Вами.
Послать запрос