1
0
Fork 0
mirror of https://gitlab.com/niansa/discord_llama.git synced 2025-03-06 20:48:25 +01:00
Commit graph

227 commits

Author SHA1 Message Date
niansa
7aba75f7a2 Implemented parallel inference 2023-05-05 21:16:19 +02:00
niansa
a6be845652 Revert "Initial CoSched support"
This reverts commit be9828adea
2023-05-04 13:26:40 +00:00
be9828adea Initial CoSched support 2023-05-04 15:24:06 +02:00
niansa
f9196ed18a Fixed this_shard sqlite select 2023-05-02 19:55:59 +02:00
niansa
259053f7e0 Make sure threads don't shift on reshard 2023-05-02 19:45:13 +02:00
niansa
20553b75a0 Fixed typo "weight" -> "weights" 2023-05-01 17:42:44 +02:00
niansa
ed01d29ef0 Reworked configuration system 2023-05-01 17:30:42 +02:00
niansa
f0d1a1c956 Fixed custom status texts not being applied 2023-04-30 00:54:36 +02:00
niansa
62fe173b77 Made status texts configurable 2023-04-30 00:52:12 +02:00
niansa
db7351a6c8 Run cleanup periodically 2023-04-30 00:36:47 +02:00
niansa
a0c4e9d856 Don't attempt to delete thread creation messaage from all embeds 2023-04-30 00:04:15 +02:00
niansa
d60a1eebaa Implemented max context age 2023-04-29 23:52:40 +02:00
niansa
18b1acb00e Don't use exception to trigger stop 2023-04-29 23:14:24 +02:00
niansa
35ecc70d45 Only attempt to stop() the Bot on main thread 2023-04-29 23:08:23 +02:00
niansa
53b092f40a Added signal handling for Linux 2023-04-29 23:04:13 +02:00
niansa
1f74d4ee54 Don't break on broken utf-8 2023-04-29 22:28:51 +02:00
niansa
915902ec4d Display first message in embed as quote 2023-04-29 22:00:33 +02:00
niansa
1611b82e37 Improved max_words utils function 2023-04-29 21:47:32 +02:00
niansa
2addd816e2 Only show first message in thread embed 2023-04-29 21:40:23 +02:00
niansa
6f1fae7a7f Fixed link in embed 2023-04-29 21:35:25 +02:00
niansa
dfcdcc7c1b Reorganization and added experimental thread embeds 2023-04-29 21:26:13 +02:00
niansa
f113ae5a81 Removed message length limit, we have a timeout now 2023-04-29 14:38:57 +02:00
niansa
3379a528f7 Make reached timeout clearer 2023-04-29 13:03:56 +02:00
niansa
ae9f11dd02 Show shard ID in thread name 2023-04-28 23:18:05 +02:00
niansa
85f2777f0e Added mutex to command completion buffer 2023-04-28 23:13:35 +02:00
niansa
ca5e3a64f4 In on_own_shard, use creation time as source instead 2023-04-28 22:59:51 +02:00
niansa
356611cb83 Fixed segfault on thread creation 2023-04-28 22:52:59 +02:00
niansa
a4faad147c In on_own_shard, divide ID by 10 to get more even results 2023-04-28 22:48:56 +02:00
niansa
1b2c5c14c5 Improved log messages 2023-04-28 22:48:32 +02:00
niansa
aa0db8ba8b Allow slash commands to be sharded 2023-04-28 22:37:41 +02:00
niansa
38862ce5fb Improved sharding 2023-04-28 21:20:37 +02:00
fd38528dc2 Fixed typo 'cout' -> 'count' 2023-04-28 18:58:18 +00:00
niansa
31885f6cd2 Implemented proper scrolling 2023-04-28 18:04:53 +02:00
niansa
524d90e138 Added simple sharding 2023-04-28 10:31:55 +02:00
niansa
18e953d850 Made instruct_mode parameter default to true 2023-04-28 10:25:13 +02:00
f677ef59ab Fixed translation with no_translate enabled 2023-04-28 00:22:01 +02:00
8706839521 Fixed crash when threads model is not found 2023-04-28 00:13:37 +02:00
niansa
6eec277861 Do inference reset in llm thread 2023-04-27 20:06:55 +02:00
niansa
395e76fdea Added !reset command 2023-04-27 20:02:36 +02:00
niansa
489d8d1704 Fixed inability to find models with certain names 2023-04-27 19:26:18 +02:00
b1445ba111 Removed debug store command 2023-04-27 14:41:54 +02:00
771cf8b785 Improved multi-line instruct mode 2023-04-27 14:41:19 +02:00
46480005d4 Added timeout config key 2023-04-27 13:47:18 +02:00
d3aff11b82 Don't catch context length exceptions
...as they are no longer thrown
2023-04-27 10:06:26 +02:00
3979b4efc1 Fixed compiler warning 2023-04-27 08:38:57 +02:00
niansa
301bb529a5 Don't delete message on timeout in live edit mode 2023-04-25 21:17:09 +02:00
niansa
d824995389 Fixed non-instruct mode 2023-04-25 21:14:45 +02:00
niansa
e7aff482fc Added no_translate model config option 2023-04-25 19:47:52 +02:00
niansa
9cead67daf Clarify what prompt failed to open 2023-04-25 18:23:16 +02:00
niansa
123cb0fad0 Fixed warning about live edit 2023-04-25 18:16:14 +02:00