Alpa is a system for training large-scale neural networks. Scaling neural networks to hundreds of billions of parameters has enabled dramatic breakthroughs such as GPT-3, but training these large-scale neural networks requires complicated distributed training techniques. Alpa aims to automate large-scale distributed training with just a few lines of code.
The key features of Alpa include:
💻 Automatic Parallelization. Alpa automatically parallelizes users' single-device code on distributed clusters with data, operator, and pipeline parallelism.
🚀 Excellent Performance. Alpa achieves linear scaling on training models with billions of parameters on distributed clusters.
✨ Tight Integration with Machine Learning Ecosystem. Alpa is backed by open-source, high-performance, and production-ready libraries such as Jax, XLA, and Ray
Use Alpa's decorator @parallelize to scale your single-device training code to distributed clusters.
import alpa
# Parallelize the training step in Jax by simply using a decorator
@alpa.parallelize
def train_step(model_state, batch):
def loss_func(params):
out = model_state.forward(params, batch["x"])
return jnp.mean((out - batch["y"]) ** 2)
grads = grad(loss_func)(model_state.params)
new_model_state = model_state.apply_gradient(grads)
return new_model_state
# The training loop now automatically runs on your designated cluster
model_state = create_train_state()
for batch in data_loader:
model_state = train_step(model_state, batch)Check out the Alpa Documentation site for installation instructions, tutorials, examples, and more.
- Alpa paper (OSDI'22)
- Google AI blog
- Alpa talk slides
- Big model tutorial (ICML'22)
- Please read the contributor guide if you are interested in contributing to Alpa.
- Connect to Alpa contributors via the Alpa slack.
Alpa is licensed under the Apache-2.0 license.