Software Development Kit
GTI’s Software Development Kit provides a hardware-accelerated, Convolutional Neural Network (CNN) system and supporting software library – implementing state-of-the-art algorithms in fast hardware on the Lightspeeur® series of AI accelerators.
SDK 5.x requires a special model structure introduced in the following MDKs (Caffe 22.214.171.124 and TensorFlow 126.96.36.199).
- SDK for Linux x86 Platform (version 188.8.131.52) – 205 MB
- SDK for ARMv7 (version 184.108.40.206) – 146 MB
- SDK for AArch64 (version 220.127.116.11) – 134 MB
- SDK for Windows (version 18.104.22.168) - 161 MB
- SDK for Linux x86 (version 22.214.171.124) – 887 MB
- SDK for ARMv7 (version 126.96.36.199) – 775 MB
- SDK for AArch64 (version 188.8.131.52) – 773 MB
- SDK for Windows (version 184.108.40.206) – 804 MB
PlaiTM Builder as a PyTorch based “Full-Stack” framework specially designed to quickly build and train neural network models for Lightspeeur® solutions.
Once you download Plai™ Builder and follow the installation instructions described within the User Guide, you can choose between GUI and command line options.
- Plai Builder v0.1 – (700MB) — Works with 2801 only
Model Development Kits (MDK) – PREMIUM PAID MEMBERS ONLY
GTI’s Model Development Kits (MDKs) enable customers to build and train neural network models specifically for GTI AI accelerator chips.
Both Caffe and TensorFlow are fully supported frameworks.
These kits are used to port neural networks onto GTI devices with comparable performance to the original floating-point model. Train models from scratch or fine-tune from a pre-trained floating-point model. Working examples of VGG16, ResNet18 and MobileNet v1 are provided.
Additionally, the conversion tool optimizes models for use with GTI’s Lightspeeur® series of accelerator chips.
Models developed with the latest MDKs are compatible with SDK 5.x only
In order for models developed with the following MDKs to be compatible with SDK 5.x, they must be converted with updateModelTool inside SDK 5.0