mirror of
https://github.com/deepspeedai/DeepSpeed.git
synced 2025-10-20 15:33:51 +08:00
227a60c0c4
This PR introduces *DeepCompile*, a new feature that efficiently integrates compiler optimizations with other DeepSpeed features. DeepCompile utilizes torch's dynamo to capture the computation graph and modifies it to incorporate DeepSpeed’s optimizations seamlessly. Currently, DeepCompile supports ZeRO-1 and ZeRO-3, with enhancements such as proactive prefetching and selective unsharding to improve performance. (More details will be added later.) --------- Signed-off-by: Masahiro Tanaka <mtanaka@microsoft.com> Signed-off-by: Olatunji Ruwase <olruwase@microsoft.com> Co-authored-by: zafarsadiq <zafarsadiq120@gmail.com> Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com> Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com>
24 lines
957 B
Python
24 lines
957 B
Python
# Copyright (c) Microsoft Corporation.
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
|
|
# DeepSpeed Team
|
|
|
|
from typing import List, Tuple
|
|
from dataclasses import dataclass, field
|
|
|
|
from torch.fx import Graph
|
|
|
|
|
|
@dataclass
|
|
class ProfilingResult:
|
|
fwd_graph: Graph = None
|
|
bwd_graph: Graph = None
|
|
needs_backward: bool = False
|
|
fwd_mem: List[Tuple[str, int, int, int]] = field(default_factory=list) # name, current_alloc, delta, peak
|
|
bwd_mem: List[Tuple[str, int, int, int]] = field(default_factory=list)
|
|
fwd_time: List[Tuple[str, int, int]] = field(default_factory=list) # name, device_time, wall_time
|
|
bwd_time: List[Tuple[str, int, int]] = field(default_factory=list)
|
|
fwd_tensor_sizes: List[Tuple[str, int]] = field(default_factory=list) # name, size
|
|
bwd_tensor_sizes: List[Tuple[str, int]] = field(default_factory=list)
|
|
param_indices: List[Tuple[int, int, Tuple[int, ...]]] = field(default_factory=list) # index, ds_id, ds_shape
|