 Max Ryabinin
|
62652e1717
Add Switch Transformers-like RemoteMixtureOfExperts (#228)
|
%!s(int64=4) %!d(string=hai) anos |
 justheuristic
|
3d6a242e30
Ensure version-consistent result rounding in load_balance_peers (#230)
|
%!s(int64=4) %!d(string=hai) anos |
 Roman Zhytar
|
8c3bd93e87
Statistics averaging (#229)
|
%!s(int64=4) %!d(string=hai) anos |
 Vsevolod-pl
|
91d17a4ebc
Delta gradients transmission (#225)
|
%!s(int64=4) %!d(string=hai) anos |
 romakail
|
ca5c7610ae
Add tool for custom user experts (#189)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
32b87bf3fe
Reset gradient buffers when synchronizing with peers (#222)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
b906ae94ed
better zero_grad behavior in CollaborativeOptimizer (#221)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
2359906253
Add gradient buffers to CollaborativeOptimizer (#220)
|
%!s(int64=4) %!d(string=hai) anos |
 mponty
|
0080028e25
Add uniform compression (#202)
|
%!s(int64=4) %!d(string=hai) anos |
 ploshkin
|
9d2a40714c
Prevent DecentralizedSGD from accidentally skipping a fraction of training batches (#218)
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
916c3db52d
Move compression-related code to hivemind.utils.compression (#213)
|
%!s(int64=4) %!d(string=hai) anos |
 Alexey Bukhtiyarov
|
7bb6565674
Add CollaborativeOptimizer, TrainingAverager (#215)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
053c7c7d13
Disentangle DecentralizedAverager components, add weights (#217)
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
ca6d87a837
Replace FeedforwardBlock with a correct implementation (#211)
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
1d364b7c32
Convert SerializerBase to an abstract class (#212)
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
6128cbbd51
Add gradient clipping support to ExpertBackend (#214)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
ca3aadb8f8
Implement basic decentralized optimizers (#210)
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
6f8f192150
Improve Runtime exception handling (#207)
|
%!s(int64=4) %!d(string=hai) anos |
 Aleksandr Borzunov
|
1deab01c71
Add RSA signature protection for DHT records (#187)
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
3024d381c5
Support learning rate schedulers in ExpertBackend (#196)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
f132294edb
Extract expert-specific methods from DHT (#192)
|
%!s(int64=4) %!d(string=hai) anos |
|
Vsevolod-pl
|
b9c02ac191
Add quantile compression (#182)
|
%!s(int64=4) %!d(string=hai) anos |
 foksly
|
edf9327e45
Support client-only participants in AllReduceProtocol (#176)
|
%!s(int64=4) %!d(string=hai) anos |
|
Alexey Bukhtiyarov
|
bb7c4d5953
Fix scalar deserialization (#190)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
faa0df5b46
Cast environment variables to correct types
|
%!s(int64=4) %!d(string=hai) anos |
|
Max Ryabinin
|
bbcbd786e5
Move CLI server launch script to hivemind/hivemind_cli (#173)
|
%!s(int64=4) %!d(string=hai) anos |
|
Alexey Bukhtiyarov
|
06162992fa
Serialize DHTID source with msgpack (#172)
|
%!s(int64=4) %!d(string=hai) anos |
|
romakail
|
5a1f2a24b2
Fix a typo in quickstart.md (#174)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
23e655640c
Handle edge cases in DecentralizedAverager (#171)
|
%!s(int64=4) %!d(string=hai) anos |
|
justheuristic
|
25fdf0d94f
Move Averager metadata serialization out of user scope (#168)
|
%!s(int64=4) %!d(string=hai) anos |
