Alpa automates model-parallel training of large deep learning (DL) models by generating execution plans that unify data, operator, and pipeline parallelism. Alpa makes it simple to train and serve large models like GPT-3. Checkout the tutorials here.


TVM is an open deep learning compiler stack for cpu, gpu and specialized accelerators.
I lead the effort to generate high-performance programs with machine learning techniques. Check out the tutorials on template-based autotvm and template-free auto-scheduler (a.k.a Ansor).


MAgent is a many-agent reinforcement learning platform.
It aims at taking the number of agents in deep reinforcment learning into million scale.

Compiler Malic

Malic is the winner compiler (1/25) in Compiler 2017 at ACM Class.
It compiles a java-like language to x86-64 asm with basic optmization techniques.


WSC is a C-minus compiler runs on the Casio fx-9860 graphing calculator.
FVM is a virtual machine to provide the runtime for the bytecode generated by WSC.
With WSC & FVM, you can compile C-minus programs direcly on a Casio fx-9860 calculator.

alt text alt text