Distributed training with Keras 3
Author: Qianli Zhu
Date created: 2023/11/07
Last modified: 2023/11/07
Description: Complete guide to the distribution API for multi-backend Keras.
Introduction
The Keras distribution API is a new interface designed to facilitate
distributed deep learning across a variety of backends like JAX, TensorFlow and
PyTorch. This powerful API introduces a suite of tools enabling data and model
parallelism, allowing for efficient scaling of deep learning models on multiple
accelerators and hosts. Whether leveraging the power of GPUs or TPUs, the API
provides a streamlined approach to initializing distributed environments,
defining device meshes, and orchestrating the layout of tensors across
computational resources. Through classes like DataParallel and
ModelParallel, it abstracts the complexity involved in parallel computation,
making it easier for developers to accelerate their machine learning
workflows.
How it works
The Keras distribution API provides a global programming model that allows developers to compose applications that operate on tensors in a global context (as if working with a single device) while automatically managing distribution across many devices. The API leverages the underlying framework (e.g. JAX) to distribute the program and tensors according to the sharding directives through a procedure called single program, multiple data (SPMD) expansion.
By decoupling the application from sharding directives, the API enables running the same application on a single device, multiple devices, or even multiple clients, while preserving its global semantics.
Setup
import os
# The distribution API is only implemented for the JAX backend for now.
os.environ["KERAS_BACKEND"] = "jax"
import keras
from keras import layers
import jax
import numpy as np
from tensorflow import data as tf_data # For dataset input.
DeviceMesh and TensorLayout
The keras.distribution.DeviceMesh class in Keras distribution API represents a cluster of
computational devices configured for distributed computation. It aligns with
similar concepts in jax.sharding.Mesh and
tf.dtensor.Mesh,
where it's used to map the physical devices to a logical mesh structure.
The TensorLayout class then specifies how tensors are distributed across the
DeviceMesh, detailing the sharding of tensors along specified axes that
correspond to the names of the axes in the DeviceMesh.
You can find more detailed concept explainers in the TensorFlow DTensor guide.
# Retrieve the local available gpu devices.
devices = jax.devices("gpu") # Assume it has 8 local GPUs.
# Define a 2x4 device mesh with data and model parallel axes
mesh = keras.distribution.DeviceMesh(
shape=(2, 4), axis_names=["data", "model"], devices=devices
)
# A 2D layout, which describes how a tensor is distributed across the
# mesh. The layout can be visualized as a 2D grid with "model" as rows and
# "data" as columns, and it is a [4, 2] grid when it mapped to the physical
# devices on the mesh.
layout_2d = keras.distribution.TensorLayout(axes=("model", "data"), device_mesh=mesh)
# A 4D layout which could be used for data parallel of a image input.
replicated_layout_4d = keras.distribution.TensorLayout(
axes=("data", None, None, None), device_mesh=mesh
)
Distribution
The Distribution class in Keras serves as a foundational abstract class designed
for developing custom distribution strategies. It encapsulates the core logic
needed to distribute a model's variables, input data, and intermediate
computations across a device mesh. As an end user, you won't have to interact
directly with this class, but its subclasses like DataParallel or
ModelParallel.
DataParallel
The DataParallel class in the Keras distribution API is designed for the
data parallelism strategy in distributed training, where the model weights are
replicated across all devices in the DeviceMesh, and each device processes a
portion of the input data.
Here is a sample usage of this class.
# Create DataParallel with list of devices.
# As a shortcut, the devices can be skipped,
# and Keras will detect all local available devices.
# E.g. data_parallel = DataParallel()
data_parallel = keras.distribution.DataParallel(devices=devices)
# Or you can choose to create DataParallel with a 1D `DeviceMesh`.
mesh_1d = keras.distribution.DeviceMesh(
shape=(8,), axis_names=["data"], devices=devices
)
data_parallel = keras.distribution.DataParallel(device_mesh=mesh_1d)
inputs = np.random.normal(size=(128, 28, 28, 1))
labels = np.random.normal(size=(128, 10))
dataset = tf_data.Dataset.from_tensor_slices((inputs, labels)).batch(16)
# Set the global distribution.
keras.distribution.set_distribution(data_parallel)
# Note that all the model weights from here on are replicated to
# all the devices of the `DeviceMesh`. This includes the RNG
# state, optimizer states, metrics, etc. The dataset fed into `model.fit` or
# `model.evaluate` will be split evenly on the batch dimension, and sent to
# all the devices. You don't have to do any manual aggregration of losses,
# since all the computation happens in a global context.
inputs = layers.Input(shape=(28, 28, 1))
y = layers.Flatten()(inputs)
y = layers.Dense(units=200, use_bias=False, activation="relu")(y)
y = layers.Dropout(0.4)(y)
y = layers.Dense(units=10, activation="softmax")(y)
model = keras.Model(inputs=inputs, outputs=y)
model.compile(loss="mse")
model.fit(dataset, epochs=3)
model.evaluate(dataset)
Epoch 1/3
8/8 ━━━━━━━━━━━━━━━━━━━━ 8s 30ms/step - loss: 1.0116
Epoch 2/3
8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 0.9237
Epoch 3/3
8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 0.8736
8/8 ━━━━━━━━━━━━━━━━━━━━ 5s 5ms/step - loss: 0.8349
0.842325747013092
ModelParallel and LayoutMap
ModelParallel will be mostly useful when model weights are too large to fit
on a single accelerator. This setting allows you to spit your model weights or
activation tensors across all the devices on the DeviceMesh, and enable the
horizontal scaling for the large models.
Unlike the DataParallel model where all weights are fully replicated,
the weights layout under ModelParallel usually need some customization for
best performances. We introduce LayoutMap to let you specify the
TensorLayout for any weights and intermediate tensors from global perspective.
LayoutMap is a dict-like object that maps a string to TensorLayout
instances. It behaves differently from a normal Python dict in that the string
key is treated as a regex when retrieving the value. The class allows you to
define the naming schema of TensorLayout and then retrieve the corresponding
TensorLayout instance. Typically, the key used to query
is the variable.path attribute, which is the identifier of the variable.
As a shortcut, a tuple or list of axis
names is also allowed when inserting a value, and it will be converted to
TensorLayout.
The LayoutMap can also optionally contain a DeviceMesh to populate the
TensorLayout.device_mesh if it is not set. When retrieving a layout with a
key, and if there isn't an exact match, all existing keys in the layout map will
be treated as regex and matched against the input key again. If there are
multiple matches, a ValueError is raised. If no matches are found, None is
returned.
mesh_2d = keras.distribution.DeviceMesh(
shape=(2, 4), axis_names=["data", "model"], devices=devices
)
layout_map = keras.distribution.LayoutMap(mesh_2d)
# The rule below means that for any weights that match with d1/kernel, it
# will be sharded with model dimensions (4 devices), same for the d1/bias.
# All other weights will be fully replicated.
layout_map["d1/kernel"] = (None, "model")
layout_map["d1/bias"] = ("model",)
# You can also set the layout for the layer output like
layout_map["d2/output"] = ("data", None)
model_parallel = keras.distribution.ModelParallel(
mesh_2d, layout_map, batch_dim_name="data"
)
keras.distribution.set_distribution(model_parallel)
inputs = layers.Input(shape=(28, 28, 1))
y = layers.Flatten()(inputs)
y = layers.Dense(units=200, use_bias=False, activation="relu", name="d1")(y)
y = layers.Dropout(0.4)(y)
y = layers.Dense(units=10, activation="softmax", name="d2")(y)
model = keras.Model(inputs=inputs, outputs=y)
# The data will be sharded across the "data" dimension of the method, which
# has 2 devices.
model.compile(loss="mse")
model.fit(dataset, epochs=3)
model.evaluate(dataset)
Epoch 1/3
/opt/conda/envs/keras-jax/lib/python3.10/site-packages/jax/_src/interpreters/mlir.py:761: UserWarning: Some donated buffers were not usable: ShapedArray(float32[784,50]).
See an explanation at https://jax.readthedocs.io/en/latest/faq.html#buffer-donation.
warnings.warn("Some donated buffers were not usable:"
8/8 ━━━━━━━━━━━━━━━━━━━━ 5s 8ms/step - loss: 1.0266
Epoch 2/3
8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - loss: 0.9181
Epoch 3/3
8/8 ━━━━━━━━━━━━━━━━━━━━ 0s 5ms/step - loss: 0.8725
8/8 ━━━━━━━━━━━━━━━━━━━━ 1s 4ms/step - loss: 0.8381
0.8502610325813293
It is also easy to change the mesh structure to tune the computation between more data parallel or model parallel. You can do this by adjusting the shape of the mesh. And no changes are needed for any other code.
full_data_parallel_mesh = keras.distribution.DeviceMesh(
shape=(8, 1), axis_names=["data", "model"], devices=devices
)
more_data_parallel_mesh = keras.distribution.DeviceMesh(
shape=(4, 2), axis_names=["data", "model"], devices=devices
)
more_model_parallel_mesh = keras.distribution.DeviceMesh(
shape=(2, 4), axis_names=["data", "model"], devices=devices
)
full_model_parallel_mesh = keras.distribution.DeviceMesh(
shape=(1, 8), axis_names=["data", "model"], devices=devices
)