59 commits

Author	SHA1	Message	Date
niansa	90e54d66d0	Removed cosched support	2024-03-25 01:18:37 +01:00
niansa	fc5e4f5aa1	Updated llama.cpp-mainline	2023-10-04 22:13:48 +02:00
niansa	79cf49faae	Implemented grammar sampling and zero-temperature sampling	2023-08-31 19:37:33 +02:00
niansa	01b0d059ed	Added pre_tick	2023-06-15 18:14:09 +02:00
niansa	0199db02b7	Added GPU support	2023-06-10 00:49:21 +02:00
niansa	94953cd174	Improve some error handling macros	2023-06-09 23:53:01 +02:00
niansa	53a4623aef	Added mirostat support	2023-05-26 00:43:07 +02:00
niansa	5feca59be7	Fixed linebreaks and support latest llama.cpp	2023-05-20 02:25:46 +02:00
niansa	a98784aa53	Minor MPT improvements	2023-05-16 23:35:42 +02:00
niansa	60fe6b9c55	Load implemenations as shared objects	2023-05-16 19:10:05 +00:00
niansa	f3a9092ca5	Moved gptj/utils.* to g4a-common.*	2023-05-15 09:44:04 +02:00
niansa	36a6fa6290	Warn about use of exceptions with cosched properly	2023-05-15 09:40:41 +02:00
niansa	81c8f39732	Added missing LM_COTHROW define	2023-05-15 09:38:49 +02:00
niansa	087fe1396b	Fixed all other known compilation issues	2023-05-10 21:50:37 +02:00
niansa	b61c751d33	Reverted last commit. but fixed invalid ssize_t typedef on msvc	2023-05-10 16:39:44 +02:00
niansa	bdb87534e8	Eliminate use of ssize_t	2023-05-10 16:37:43 +02:00
niansa	0f8ef2fd32	Removed if_error because it wasn't needed	2023-05-10 12:18:42 +02:00
niansa	c9d261701d	(Hopefully) fixed msvc compilation	2023-05-10 12:14:49 +02:00
niansa	6968f3459a	Made last commit more beautiful	2023-05-09 22:25:04 +02:00
niansa	5e666d83db	Actually implemented error return values	2023-05-09 22:20:01 +02:00
niansa	0d5cba0530	Added LM_NOEXCEPT cmake option	2023-05-09 21:33:49 +02:00
niansa	ca33a27e05	Fixed a warning	2023-05-06 01:14:24 +02:00
niansa	7076f863d4	Check for task termination	2023-05-05 19:02:28 +02:00
niansa	96076d4e9b	Updated for latest cosched	2023-05-05 10:35:41 +02:00
niansa	58b3a278d8	Don't use weak pointers	2023-05-04 15:53:19 +02:00
niansa	5a57db5e75	Added CoSched support	2023-05-04 15:22:32 +02:00
niansa	3c62cbebaf	Inlined InferencePool constructor and added addional cleanup functions	2023-04-29 23:43:45 +02:00
niansa	d236e36d26	Implemented proper scrolling	2023-04-28 18:04:07 +02:00
niansa	17f30e0761	Removed uninitialized GPTJInference::State::n_ctx	2023-04-27 10:41:56 +02:00
niansa	e640f093df	Removed context size exception	2023-04-27 10:04:03 +02:00
niansa	493186509a	Renamed function and updated Python bindings	2023-04-27 09:48:44 +02:00
niansa	94e4ca5874	Set default context size limit	2023-04-27 09:47:23 +02:00
niansa	ca4ad5f096	Added context window scrolling with top bar	2023-04-27 09:45:37 +02:00
niansa	219186f4b6	Take const string reference instead of string view in append()	2023-04-27 09:31:22 +02:00
niansa	4e74517bb5	Changed parameter types to some that make more sense	2023-04-27 09:27:09 +02:00
niansa	0661b2e33d	Updated for latest llama.cpp and working gpt-j implementation	2023-04-27 08:21:02 +02:00
niansa	5f6cf17871	Moved pool functions into seperate file	2023-04-26 11:18:12 +02:00
niansa	1f75673523	Fixed compilation	2023-04-26 11:10:25 +02:00
niansa	aad1bd9ae4	Made Inference class virtual	2023-04-26 10:59:24 +02:00
niansa	55a310b005	Synced top_p default with llama.cpp	2023-04-25 17:08:55 +02:00
niansa	d88dc5ad98	Fixed InferencePool::store_all() storing empty slots	2023-04-25 15:13:57 +02:00
niansa	97f7f2ebc5	Removed debugging left-over	2023-04-23 18:24:42 +02:00
niansa	138a9bde52	Minor fixes and Python bindings	2023-04-23 18:06:57 +02:00
niansa	0466774286	Fully implemented InferencePool	2023-04-23 15:31:16 +02:00
niansa	414554c69a	Use ios::binary for file streams	2023-04-23 12:40:29 +02:00
niansa	cf2fec84e0	Added experimental InferencePool	2023-04-23 12:31:39 +02:00
niansa	e85556d94d	Added Inference::get_prompt	2023-04-20 21:34:20 +02:00
niansa	677ac470e7	Load/Restore prompt	2023-04-20 21:32:45 +02:00
niansa	06d99a6950	Allow generation without end string and fixed Python bug	2023-04-18 01:17:32 +02:00
niansa	80d33df458	Added is_valid to Savestate	2023-04-17 22:55:28 +02:00