 Alexander Borzunov
|
0b3dcf4da0
Set default DHT num_workers = 4 (#342)
|
4 жил өмнө |
|
Alexander Borzunov
|
3f691fced4
Convert averager to libp2p backend (#323)
|
4 жил өмнө |
 justheuristic
|
11db5fd56f
Refactor for v0.9.10 and fix example (#319)
|
4 жил өмнө |
 Max Ryabinin
|
2f07a556e6
Reformat code with Black (#274)
|
4 жил өмнө |
|
Alexander Borzunov
|
0be1512c74
Convert DHT to libp2p backend (#296)
|
4 жил өмнө |
|
Max Ryabinin
|
5233b6c085
Split hivemind.client into hivemind.averaging and hivemind.moe (#304)
|
4 жил өмнө |
|
justheuristic
|
200fbecdbf
Refactor MPFuture to use a single pipe/thread per process (#298)
|
4 жил өмнө |
|
justheuristic
|
e9956b84f6
Add graceful shutdown to DHT and Averager (#301)
|
4 жил өмнө |
|
justheuristic
|
f0c5627139
Improve error handling, remove deprecated functionality (#261)
|
4 жил өмнө |
|
Max Ryabinin
|
2328ba9262
Fix device in Switch-MoE, overhaul Server architecture (#256)
|
4 жил өмнө |
|
Aleksandr Borzunov
|
18add2c04b
Implement combining validators (#249)
|
4 жил өмнө |
|
Max Ryabinin
|
62652e1717
Add Switch Transformers-like RemoteMixtureOfExperts (#228)
|
4 жил өмнө |
|
justheuristic
|
f132294edb
Extract expert-specific methods from DHT (#192)
|
4 жил өмнө |
|
justheuristic
|
1d1252c30d
Load state from peers in DecentralizedAverager (#154)
|
4 жил өмнө |
|
justheuristic
|
10917b259e
Averager: update group keys after every step, infer nbits dynamically (#141)
|
4 жил өмнө |
|
justheuristic
|
c36b5b1a9b
Add DHT peer validation, add DHT.get_visible_address, add blacklist for unresponsive peers (#137)
|
4 жил өмнө |
|
justheuristic
|
e159605143
Address averaging corner cases, add benchmark_averaging.py, chunk averaged tensors, fix DHTNode get (#134)
|
4 жил өмнө |
|
Max Ryabinin
|
d6ac1fbd8a
Remove duplicate log entries, report aggregate runtime performance and parameter count (#135)
|
4 жил өмнө |
|
justheuristic
|
eb93789ac6
Implement averaging parameters over DHT (2/3)
|
4 жил өмнө |
|
Max Ryabinin
|
34585bafac
DHT caching and TransformerEncoderLayer fixes (#124)
|
4 жил өмнө |
|
justheuristic
|
1754792aad
Process-wide channel cache for gRPC+aio (#120)
|
4 жил өмнө |
|
justheuristic
|
2bd481c73f
added torch1.7 support, switch to grpc 1.33, grpc bump, improved tests & logging, (#116)
|
4 жил өмнө |
|
justheuristic
|
8e068a8341
Negative cache for DHT (#114)
|
4 жил өмнө |
|
justheuristic
|
621ac4c0e9
Switch to new beam search in client (#111)
|
4 жил өмнө |
|
justheuristic
|
a59fa709cc
Use namedtuples for DHT values (#110)
|
4 жил өмнө |
|
Max Ryabinin
|
70dadfb8b5
Start the server with a terminal command, fix default config.yml (#108)
|
4 жил өмнө |
|
justheuristic
|
9f9c4ac96b
Faster beam search through DHT sub-keys (#107)
|
4 жил өмнө |
|
justheuristic
|
9c1e14aca1
Faster beam search: part1/2 (#97)
|
5 жил өмнө |
|
Max Ryabinin
|
53f5ab6971
Speed up tests, shutdown threads via Event (#82)
|
5 жил өмнө |
|
justheuristic
|
6d2b8094c9
RemoteMixtureOfExperts update part II (#80)
|
5 жил өмнө |
