Kernels:
Trusted publisher
File size: 1,790 Bytes
64ea34e | 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 49 50 51 | # Copyright (c) 2023-2026, Songlin Yang, Yu Zhang, Zhiyuan Li
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
# For a list of all contributors, visit:
# https://github.com/fla-org/flash-linear-attention/graphs/contributors
import torch.nn as nn
from torch.distributed import DeviceMesh
from torch.distributed.tensor import Replicate, distribute_module
from torch.distributed.tensor.parallel import ParallelStyle
from torch.distributed.tensor.placement_types import Placement
try:
from torch.distributed.tensor import DTensor
except (ImportError, AttributeError):
DTensor = None
class PrepareModuleWeight(ParallelStyle):
"""Wrap parameters that already match their layout on a one-dimensional mesh.
Replicated values must match across ranks; sharded values must already be local slices.
"""
def __init__(self, *, layouts: Placement | None = None):
super().__init__()
self.layouts = Replicate() if layouts is None else layouts
def _replicate_module_fn(
self,
name: str,
module: nn.Module,
device_mesh: DeviceMesh,
):
for p_name, param in module.named_parameters(recurse=False):
replicated_param = nn.Parameter(
DTensor.from_local(param, device_mesh, [self.layouts], run_check=False),
requires_grad=param.requires_grad,
)
module.register_parameter(p_name, replicated_param)
def _apply(self, module: nn.Module, device_mesh: DeviceMesh) -> nn.Module:
return distribute_module(
module,
device_mesh,
partition_fn=self._replicate_module_fn,
input_fn=None,
output_fn=None,
)
|