Max Ryabinin
|
0fc36c8eed
Add standard experts
|
2 жил өмнө |
Max Ryabinin
|
03ebe9c6c6
Add standard experts
|
2 жил өмнө |
Max Ryabinin
|
28f5702edf
Forbid protobuf>=4.0
|
2 жил өмнө |
Max Ryabinin
|
77919315c3
Revert "Test bilevel queue"
|
3 жил өмнө |
Max Ryabinin
|
ff7f44b01e
Revert "Test bilevel queue"
|
3 жил өмнө |
Max Ryabinin
|
e17610f8cf
Revert "No 1.5 multiplier"
|
3 жил өмнө |
Max Ryabinin
|
eb06507018
No 1.5 multiplier
|
3 жил өмнө |
Max Ryabinin
|
7e6bdc14bd
Test bilevel queue
|
3 жил өмнө |
Max Ryabinin
|
11848f6bc1
Test bilevel queue
|
3 жил өмнө |
Max Ryabinin
|
339f35f25d
Extra logging
|
3 жил өмнө |
Max Ryabinin
|
9efff5fa6f
pass -> raise
|
3 жил өмнө |
Max Ryabinin
|
d5bf507ff6
Don't ban experts for timeout
|
3 жил өмнө |
Max Ryabinin
|
b9ccbe7b48
Don't ban experts for timeout
|
3 жил өмнө |
Max Ryabinin
|
1cfd86ac5b
Add timeouts, remove gated for tests
|
3 жил өмнө |
Max Ryabinin
|
b48220577e
Add server-side gradient accumulation
|
3 жил өмнө |
Max Ryabinin
|
b26d61b1c4
Add optional offload
|
3 жил өмнө |
Max Ryabinin
|
9b5ee08bd6
Support FP16
|
3 жил өмнө |
Max Ryabinin
|
a3918cd063
Reduce DHTHandler metadata storage time
|
3 жил өмнө |
Max Ryabinin
|
6ca02e21cf
Increase DHTExpiration
|
3 жил өмнө |
Max Ryabinin
|
40662e3800
Increase timeout
|
3 жил өмнө |
Max Ryabinin
|
66e387d23c
Increase timeout
|
3 жил өмнө |
Max Ryabinin
|
d800ff438e
Increase compression
|
3 жил өмнө |
Max Ryabinin
|
7ffab7740c
Initialize scheduler correctly
|
3 жил өмнө |
Max Ryabinin
|
61f20884b9
Attempt training without offload
|
3 жил өмнө |
Max Ryabinin
|
cf108bad0d
Extra debug prints
|
3 жил өмнө |
Max Ryabinin
|
01273b8241
Extra debug prints
|
3 жил өмнө |
Max Ryabinin
|
0ff0c689e8
Remove AMP, update lr
|
3 жил өмнө |
Max Ryabinin
|
04173527c3
hidden_act_gated = True
|
3 жил өмнө |
Max Ryabinin
|
ff826d7667
Extra collaboration prefix logging
|
3 жил өмнө |
Max Ryabinin
|
836192eadc
WIP
|
3 жил өмнө |