Skip to content

Use msgpack for p2p comm #6547

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
loadams merged 2 commits into from
Sep 26, 2024
Merged

Use msgpack for p2p comm #6547

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5abb60b
use msgpack for p2p comm
Sep 17, 2024
5654d89
Merge branch 'master' into tohtana/remove_p2p_pickle
loadams Sep 25, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions deepspeed/runtime/pipe/p2p.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@

# DeepSpeed Team

import pickle
import msgpack
import typing

import torch
Expand Down Expand Up @@ -96,7 +96,7 @@ def wait():
def send_obj(msg: typing.Any, dest: int):
"""Send an arbitrary python object to ``dest``.

Note: ``msg`` must be pickleable.
Note: ``msg`` must be serializable by msgpack.

WARN: This incurs a CPU -> GPU transfer and should be used sparingly
for performance reasons.
Expand All @@ -106,7 +106,7 @@ def send_obj(msg: typing.Any, dest: int):
dest (int): Destination rank.
"""
# serialize the message
msg = pickle.dumps(msg)
msg = msgpack.packb(msg)
# construct a tensor to send
msg = torch.ByteTensor(torch.ByteStorage.from_buffer(msg)).to(get_accelerator().device_name())

Expand All @@ -133,7 +133,7 @@ def recv_obj(sender: int) -> typing.Any:
msg = torch.empty(length.item(), dtype=torch.uint8).to(get_accelerator().device_name())
dist.recv(msg, src=sender)

msg = pickle.loads(msg.cpu().numpy().tobytes())
msg = msgpack.unpackb(msg.cpu().numpy().tobytes())

def _to(x):
"""Recursively move to the current device."""
Expand Down
1 change: 1 addition & 0 deletions requirements/requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
hjson
msgpack
ninja
numpy
packaging>=20.0
Expand Down