Skip to content
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

Jump to bottom

Add downloading time to averager #618

Merged
merged 2 commits into from
Jun 12, 2024
Merged

Add downloading time to averager #618

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
050bb63
add dowloading time to averager
samsja Jun 5, 2024
c84e6dd
fix use monotonic time
samsja Jun 10, 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
5 changes: 4 additions & 1 deletion hivemind/averaging/averager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@
import random
import signal
import threading
import time
import weakref
from dataclasses import asdict
from typing import Any, AsyncIterator, Dict, Optional, Sequence, Tuple, Union
Expand Down Expand Up @@ -78,7 +79,7 @@
local tensors for averaging
:param allow_state_sharing: if set to True, other peers can download this peer's state. Can be overwritten
with averager.allow_state_sharing = True / False
:param declare_state_period: re-declare averager as a donor for load_state_from_peers every this many seconds

Check failure on line 82 in hivemind/averaging/averager.py

View workflow job for this annotation

GitHub Actions / codespell 

re-declare ==> redeclare
:param allreduce_timeout: spend at most this many seconds for allreduce (after group is formed)
:param next_chunk_timeout: during all-reduce and load_state_from_peers, if peer does not send next data chunk in
this number of seconds, consider it failed and proceed with remaining peers. default: no timeout
Expand Down Expand Up @@ -700,6 +701,7 @@
metadata = None
for peer in sorted(peer_priority.keys(), key=peer_priority.get, reverse=True):
if peer != self.peer_id:
t0 = time.monotonic()
logger.info(f"Downloading parameters from peer {peer}")
try:
stub = self.get_stub(self._p2p, peer, namespace=self.prefix)
Expand All @@ -722,7 +724,8 @@
logger.debug(f"Peer {peer} did not send its state")
continue

logger.info(f"Finished downloading state from {peer}")
t1 = time.monotonic()
logger.info(f"Finished downloading state in {t1 - t0:.3f}s from {peer}")
future.set_result((metadata, tensors))
return
except Exception as e:
Expand Down
Loading