https://github.com/epfml/pax
JAX-like API for PyTorch
https://github.com/epfml/pax
Last synced: 10 months ago
JSON representation
JAX-like API for PyTorch
- Host: GitHub
- URL: https://github.com/epfml/pax
- Owner: epfml
- License: apache-2.0
- Created: 2021-08-23T13:35:57.000Z (over 4 years ago)
- Default Branch: main
- Last Pushed: 2022-03-02T12:19:38.000Z (almost 4 years ago)
- Last Synced: 2025-04-18T16:15:33.599Z (11 months ago)
- Language: Python
- Homepage:
- Size: 457 KB
- Stars: 3
- Watchers: 8
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
Sharing some of JAX's beautiful API with PyTorch users.
Disambiguation: for wardrobes, see [link](https://www.ikea.com/ch/en/cat/pax-system-19086/). For peace, you are at the right spot.
## Installation
```bash
pip install paxlib
```
or
```bash
pip install git+https://github.com/epfml/pax.git
```
## Pytrees in PyTorch
```python
import torch
import pax
tree = {
"a": [torch.tensor(3.0), torch.tensor(4.0)],
"c": 4
}
pax.tree_map(lambda x: x*2, tree)
```
Note: we currently depend on `jax` for this functionality, but we could use [dm-tree](https://github.com/deepmind/tree) instead to drop the dependency.
## Autodiff that looks like JAX
We follow the API of `jax.grad`:
```python
import pax
f = lambda x: x**2
df_dx = pax.grad(f)
df_dx(2.0) # tensor(4.0)
```
This works with any Pytree as input:
```python
def f(x):
return x["a"] * x["b"]
x = {"a": 2.0, "b": -1.5}
pax.value_and_grad(f)(x) # (tensor(-3.), {'a': tensor(-1.5000), 'b': tensor(2.)})
```
PAX also supports higher-order derivatives:
```python
f = lambda x: 1/6 * x**3
pax.grad(f)(2.0) # tensor(2.)
pax.grad(pax.grad(f))(3.0) # tensor(3.)
```
## Example: Minimal SGD
```python
import torch
import pax
f = lambda x: x**2
df_dx = pax.grad(f)
x = torch.randn([]) # initialization
for step in range(20):
x = x - 0.1 * df_dx(x)
print(x, f(x))
```
## Example: meta-learning the learning rate
```python
f = lambda x: x**2
df_dx = pax.grad(f)
def sgd(x, lr=0.1, num_steps=10):
for _ in range(num_steps):
x = x - lr * df_dx(x)
return x
# optimize the learning rate
def meta_loss(lr):
x0 = 1.0
return f(sgd(x0, lr=lr))
df_dlr = pax.grad(meta_loss)
lr = 0.1
for _ in range(100):
lr = lr - 0.1 * df_dlr(lr)
```
## Converting from PyTorch
We provide a small wrapper for PyTorch _modules_ to make them behave like [Haiku](https://github.com/deepmind/dm-haiku).
```python
net = torch.nn.Linear(10, 1) # any torch.nn.Module
# convert
forward = pax.functional_module(net)
# intialize
params, buffers = pax.get_params(net), pax.get_buffers(net)
# run
data_batch = torch.zeros(2, 10)
out, buffers = forward(params, data_batch, buffers=buffers, is_training=True)
```
and also a wrapper to make PyTorch _optimizers_ functional like [Optax](https://github.com/deepmind/optax):
```python
optimizer = pax.functional_optimizer(torch.optim.Adam, lr=1e-3)
f = lambda x: x**2
df_dx = pax.grad(f)
params = torch.tensor(3.)
opt_state = optimizer.init(params)
for step in range(10):
params, opt_state = optimizer.step(params, df_dx(params), opt_state)
print(params.item())
```
Using PAX optimizers with learning rate schedulers looks like this:
```python
optimizer = pax.functional_optimizer(torch.optim.SGD, lr=0)
lr_at_step = pax.functional_schedule(torch.optim.lr_scheduler.LambdaLR, lr_lambda=lambda step: 1/(step+1), initial_lr=0.1)
f = lambda x: x**2
df_dx = pax.grad(f)
params = torch.tensor(3.)
opt_state = optimizer.init(params)
for step in range(10):
params, opt_state = optimizer.step(params, df_dx(params), opt_state, lr=lr_at_step(step))
print(params.item())
```
## Runtime overhead
We measured the time required for one epoch on CIFAR-10 with a batch size of 128.
We compare a standard PyTorch implementation based on [this tutorial](https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html) to a PAX one, using `pax.value_and_grad`, `pax.functional_module` and `pax.functional_optimizer`. This is currently a little slower than regular PyTorch code. The peak memory usage could be larger too.
