-
Notifications
You must be signed in to change notification settings - Fork 130
/
returnn_frontend.rst
48 lines (31 loc) · 1.13 KB
/
returnn_frontend.rst
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
.. _returnn_frontend:
================
RETURNN frontend
================
This is a common interface to define your models using Python code,
very similar as in PyTorch.
This common interface supports multiple backends, namely:
* PyTorch
* TensorFlow layer dictionaries
* TensorFlow directly
Code:
.. code-block:: python
import returnn.frontend as rf
...
Related work
------------
`Ivy <https://github.com/unifyai/ivy>`__
is both an ML transpiler and a framework,
currently supporting JAX, TensorFlow, PyTorch, and Numpy.
`Keras Core <https://keras.io/keras_core/>`__:
Keras for TensorFlow, JAX and PyTorch.
Also can wrap pure PyTorch modules directly in ``keras.Model``
(`example <https://twitter.com/fchollet/status/1697381832164290754>`__).
`PyTorch-to-RETURNN <https://github.com/rwth-i6/pytorch-to-returnn>`__:
Convert PyTorch models to RETURNN TF layer dicts semi-automatically.
`JAX2Torch <https://github.com/lucidrains/jax2torch>`__:
Use Jax functions and models in PyTorch.
(`AlphaFold example <https://twitter.com/sokrypton/status/1623914503950983168>`__).
History
-------
https://github.com/rwth-i6/returnn/issues/1120