 justheuristic
|
78c59bdfbe
review
|
3 anni fa |
 Max Ryabinin
|
bb39fbdb6c
review
|
3 anni fa |
|
justheuristic
|
a9ace63365
add vision
|
3 anni fa |
|
justheuristic
|
7bd28fd3b0
add vision
|
3 anni fa |
|
justheuristic
|
033d62efdd
black-isort
|
3 anni fa |
|
justheuristic
|
500715b19e
test
|
3 anni fa |
|
justheuristic
|
f69b32a775
Merge remote-tracking branch 'origin/master' into hivemind_optimizer_thirdtimesthecharm
|
3 anni fa |
|
justheuristic
|
7a79aea439
Fix StepControl.cancel() in DecentralizedAverager (#411)
|
3 anni fa |
|
justheuristic
|
6480775b79
Merge branch 'master' into hivemind_optimizer_thirdtimesthecharm
|
3 anni fa |
|
justheuristic
|
09e34f8366
Prepare GradScaler for hivemind.Optimizer (#413)
|
3 anni fa |
|
justheuristic
|
f34d02df43
[WIP] main hivemind.Optimizer
|
3 anni fa |
|
justheuristic
|
22665fdcee
Make target group size optional (#412)
|
3 anni fa |
|
justheuristic
|
1cfc6a3b7b
Fix internal assert in GradientAverager (#410)
|
3 anni fa |
|
justheuristic
|
d883387e37
Add ProgressTracker (#408)
|
3 anni fa |
 Artem Chumachenko
|
99a0c18ca9
Catch OSError in MPFuture (#409)
|
3 anni fa |
|
justheuristic
|
02eee9292d
[hivemind.Optimizer] TrainingStateAverager (#407)
|
3 anni fa |
|
justheuristic
|
8fa0a8e6ae
Add GradientAverager with support for delayed averaging (#404)
|
3 anni fa |
|
justheuristic
|
ed4204009f
move PerformanceEMA to utils, TrainingAverager to optim, update utils (#405)
|
3 anni fa |
|
justheuristic
|
7c4d13f06d
hotfix: replace StepControl.can_modify with began_allreduce
|
3 anni fa |
|
justheuristic
|
793a741950
Move DHT to dht/dht.py, update DHT figure (#399)
|
3 anni fa |
|
justheuristic
|
a09df5492f
Add an option to pre-schedule averaging (#398)
|
3 anni fa |
|
justheuristic
|
025e095d55
backport PerformanceEMA from server_side_averaging (#397)
|
3 anni fa |
|
justheuristic
|
40d3ecebff
Fix schema typing (#396)
|
3 anni fa |
|
justheuristic
|
54cdc3925a
Apply averager updates asynchronously (#395)
|
3 anni fa |
|
justheuristic
|
09985d843b
Implement simplified all-reduce for asymmetric TCP connections (#385)
|
3 anni fa |
 Alexander Borzunov
|
91d1d31796
Fix minor issues in documentation (#392)
|
3 anni fa |
|
justheuristic
|
a02571ecb1
Hotfix codecov_in_develop_mode with --no-use-pep517
|
3 anni fa |
|
justheuristic
|
1d862c9a5d
Support different AMP & buffer configurations in one experiment, fix minor bugs (#389)
|
3 anni fa |
|
justheuristic
|
4a9bc92cd1
Implement weights as part of the allreduce protocol, not matchmaking (#384)
|
3 anni fa |
|
Alexander Borzunov
|
d809e303c5
Remove arguments with default values from example instructions (#388)
|
3 anni fa |
