jax.experimental.custom_partitioning 模块#

API#

jax.experimental.custom_partitioning.custom_partitioning(fun, static_argnums=())[源代码]#

在具有自定义 SPMD 降级规则的 XLA 图中插入 CustomCallOp。

@custom_partitioning
def f(*args):
  return ...

def propagate_user_sharding(mesh, user_shape):
  '''Update the sharding of the op from a user's shape.sharding.'''
  user_sharding = jax.tree.map(lambda x: x.sharding, user_shape)

def partition(mesh, arg_shapes, result_shape):
  def lower_fn(*args):
    ... builds computation on per-device shapes ...
  result_shardings = jax.tree.map(lambda x: x.sharding, result_shape)
  arg_shardings = jax.tree.map(lambda x: x.sharding, arg_shapes)
  # result_sharding and arg_shardings may optionally be modified and the
  # partitioner will insert collectives to reshape.
  return mesh, lower_fn, result_sharding, arg_shardings

def infer_sharding_from_operands(mesh, arg_shapes, shape):
  '''Compute the result sharding from the sharding of the operands.'''
  arg_shardings = jax.tree.map(lambda x: x.sharding, arg_shapes)


f.def_partition(partition, propagate_user_sharding, infer_sharding_from_operands)

def_partition 的参数如下

  • propagate_user_sharding: 可调用对象，它接受用户（在 dag 中）的分片，并返回新的 NamedSharding 的建议。默认实现只是返回建议的分片。

  • partition: 可调用对象，它接受 SPMD 建议的分区形状和分区规范，并返回网格、每个分片的降级函数以及最终的输入和输出分片规范（SPMD 分区器将重新分区输入以匹配）。返回网格是为了在没有提供网格时允许为集体操作配置 axis_names。

  • infer_sharding_from_operands: 可调用对象，它根据为每个参数选择的 NamedSharding 计算输出 NamedSharding。

  • decode_shardings: 设置为 True 时，如果可能，将输入 GSPMDSharding``s 转换为 ``NamedSharding。如果用户未提供上下文网格，则可能无法实现此操作。

可以使用 static_argnums 将位置参数指定为静态参数。JAX 使用 inspect.signature(fun) 来解析这些位置参数。

示例

例如，假设我们要增强现有的 jax.numpy.fft.fft。此函数计算沿最后一个维度的一维输入上的离散傅里叶变换，并沿前 N-1 个维度进行批处理。但是，默认情况下，它将忽略输入的分片并将输入收集在所有设备上。但是，由于 jax.numpy.fft.fft 沿前 N-1 个维度进行批处理，因此这是不必要的。我们将创建一个新的 my_fft 操作，该操作不会更改沿前 N-1 个维度的分片，并且仅在需要时沿最后一个维度收集输入。

import jax
from jax.sharding import NamedSharding
from jax.experimental.custom_partitioning import custom_partitioning
from jax.experimental.pjit import pjit
from jax.sharding import PartitionSpec as P
from jax.sharding import Mesh
from jax.numpy.fft import fft
import regex as re
import numpy as np

# Pattern to detect all-gather or dynamic-slice in the generated HLO
_PATTERN = '(dynamic-slice|all-gather)'

# For an N-D input, keeps sharding along the first N-1 dimensions
# but replicate along the last dimension
def supported_sharding(sharding, shape):
    rank = len(shape.shape)
    max_shared_dims = min(len(sharding.spec), rank-1)
    names = tuple(sharding.spec[:max_shared_dims]) + tuple(None for _ in range(rank - max_shared_dims))
    return NamedSharding(sharding.mesh, P(*names))

def partition(mesh, arg_shapes, result_shape):
    result_shardings = jax.tree.map(lambda x: x.sharding, result_shape)
    arg_shardings = jax.tree.map(lambda x: x.sharding, arg_shapes)
    return mesh, fft,               supported_sharding(arg_shardings[0], arg_shapes[0]),               (supported_sharding(arg_shardings[0], arg_shapes[0]),)

def infer_sharding_from_operands(mesh, arg_shapes, result_shape):
    arg_shardings = jax.tree.map(lambda x: x.sharding, arg_shapes)
    return supported_sharding(arg_shardings[0], arg_shapes[0])

@custom_partitioning
def my_fft(x):
    return fft(x)

my_fft.def_partition(
    infer_sharding_from_operands=infer_sharding_from_operands,
    partition=partition)

现在创建一个沿第一个轴分片的二维数组，将其传递给 my_fft，并注意它仍然按预期分片，并且与 fft 的输出相同。但是，检查 HLO（使用 lower(x).compile().runtime_executable().hlo_modules()）显示 my_fft 不会创建任何 all-gather 或动态切片，而 fft 会。

with Mesh(np.array(jax.devices()), ('x',)):
  x = np.asarray(np.random.randn(32*1024, 1024), dtype=np.complex64)
  y = pjit(lambda x: x, in_shardings=None, out_shardings=P('x'))(x)
  pjit_my_fft = pjit(my_fft, in_shardings=P('x'), out_shardings=P('x'))
  pjit_fft    = pjit(fft,    in_shardings=P('x'), out_shardings=P('x'))
  print(pjit_my_fft(y))
  print(pjit_fft(y))
  # dynamic-slice or all-gather are not present in the HLO for my_fft, because x is a 2D array
  assert(re.search(_PATTERN, pjit_my_fft.lower(x).compile().runtime_executable().hlo_modules()[0].to_string()) is None)
  # dynamic-slice or all-gather are present in the HLO for fft
  assert(re.search(_PATTERN, pjit_fft.lower(x).compile().runtime_executable().hlo_modules()[0].to_string())    is not None)

# my_fft
[[-38.840824   +0.j        -40.649452  +11.845365j
...
  -1.6937828  +0.8402481j  15.999859   -4.0156755j]]

# jax.numpy.fft.fft
[[-38.840824   +0.j        -40.649452  +11.845365j
  ...
  -1.6937828  +0.8402481j  15.999859   -4.0156755j]]

由于 supported_sharding 中的逻辑，my_fft 也适用于一维数组。但是，在这种情况下，my_fft 的 HLO 会显示一个动态切片，因为最后一个维度是计算 FFT 的维度，需要在计算完成之前在所有设备上进行复制。

with Mesh(np.array(jax.devices()), ('x',)):
  x = np.asarray(np.random.randn(32*1024*1024), dtype=np.complex64)
  y = pjit(lambda x: x, in_shardings=None, out_shardings=P('x'))(x)
  pjit_my_fft = pjit(my_fft, in_shardings=P('x'), out_shardings=P('x'))
  pjit_fft    = pjit(fft,    in_shardings=P('x'), out_shardings=P('x'))
  print(pjit_my_fft(y))
  print(pjit_fft(y))
  # dynamic-slice or all-gather are present in the HLO for my_fft, because x is a 1D array
  assert(re.search(_PATTERN, pjit_my_fft.lower(x).compile().runtime_executable().hlo_modules()[0].to_string()) is None)
  # dynamic-slice or all-gather are present in the HLO for fft
  assert(re.search(_PATTERN, pjit_fft.lower(x).compile().runtime_executable().hlo_modules()[0].to_string())    is not None)

# my_fft
[    7.217285   +0.j     -3012.4937  +4287.635j   -405.83594 +3042.984j
...  1422.4502  +7271.4297j  -405.84033 -3042.983j
-3012.4963  -4287.6343j]

# jax.numpy.fft.fft
[    7.217285   +0.j     -3012.4937  +4287.635j   -405.83594 +3042.984j
...  1422.4502  +7271.4297j  -405.84033 -3042.983j
-3012.4963  -4287.6343j]