NN Inference

TVM solution 

The value of open-source Apache TVM solution 

The value of TVM for Business

Speed up Time to Market 
Ready-to-run ML Compiler
Focus budget and effort on market-driven tasks
Reduction of cost on SDK development
Open-source ML Compiler under permissive Apache license
Investment’s protection of SW developments and algorithms
Only core IP development inside the team
Reduction of cost on SDK support
Updates of TVM and the training Frameworks are performed by the community

The value of TVM solution for Developers

TVM leaders’ core – high skilled toolchains experts with outstanding expertise
Community (650+ contributors, sponsored by ARM, AWS, Microsoft, Qualcomm, XILINX etc.) ensures the best implementations of enhanced optimizations and TVM components. It provides free optimizations, new frontend support, new version updates, and more
Infrastructure for automatic generation and optimization of tensor operators for vast number of backends with better performance
Heterogeneous platforms support for systems combining master processors and accelerators
Auto Scheduler (Ansor) and other tools for reducing human efforts
Easy integration on the first steps
Rapid evolution in response to current challenges and trends
Annual Apache TVM conference https://tvmconf.org/

TVM

Technicals

TVM structure and components

TVM backend types

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.

Grovety areas

YOU HAVE

WE PROPOSE

Neural network compiler
Open-source ML Compiler under permissive Apache license
Open-source ML Compiler under permissive Apache license
C/C++ / OpenCL / CUDA compiler
+ Scheduling capabilities from TVM to distribute resources on hardware modules
+ Reuse many of existing operations’ schedules from other backends as a base
+ General subgraph-level optimizations
+ Possibility to use existing proprietary optimizations and algorithms
No compiler
+ Several ways of implementing backend, number of ready-to-use backends to start from
+ Runtime implementation for various hardware types

Optional

Ready-to-run examples of several optimized NNs for Cortex M+U55/65 Dev Boards
Arm Cortex IP cores
Physical Dev Boards 
Virtual environment set-up with TVM (Docker container or VM)
Development environment for TVM 
Utility scripts for launching builds, demos, benchmarks
Dev Boards deployment at the ML CI HW Cloud Farm Service
Remote debug/CI infrastructure set-up
Remote Development Board demo access for your Customers
Contact us hi@grovety.com
Copyright © Grovety Inc.