root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-27 03:44:35 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

Commit Graph

  • a52f51ada2 Add executable bit to Miku.sh at8u 2023-04-05 16:42:45 +0100
  • c6479a3cda Define non-positive top_k; top_k range check Ivan Stepanov 2023-04-05 18:40:17 +0300
  • ed1c214e66
         zig : add build.zig (#773) iacore 2023-04-05 15:06:02 +0000
  • 0c44427df1
         make : missing host optimizations in CXXFLAGS (#763) master-0c44427 Ivan Stepanov 2023-04-05 17:38:37 +0300
  • 594cc95fab
         readme : update with CMake and windows example (#748) Adithya Balaji 2023-04-05 16:36:12 +0200
  • 88ed5761b8
         examples : add Miku.sh (#724) at8u 2023-04-05 14:32:42 +0000
  • 1868f6c84f
         ggml, llama : avoid heavy V transpose + improvements Georgi Gerganov 2023-04-05 17:04:16 +0300
  • 6f171ec28d
         Add build.zig Locria Cyber 2023-04-05 11:39:10 +0000
  • d12088e164 Minor formatting changes saharNooby 2023-04-05 15:31:23 +0400
  • 65c0af359f
         Build static lib Locria Cyber 2023-04-05 11:18:55 +0000
  • 58c438cf7d
         Add Accelerate/BLAS when using Swift (#765) Andrew Duffy 2023-04-05 11:44:24 +0100
  • 5c1920df43 why nobody ever told me the makefile doesnt work outside x86 xD Concedo 2023-04-05 17:15:42 +0800
  • 3415e292b1
         Update Package.swift Andrew Duffy 2023-04-05 09:36:28 +0100
  • 1490cdd71d change GPT-J and GPT2 KVs to use fp16 instead Concedo 2023-04-05 15:53:07 +0800
  • 86286cb318
         Fix magic in convert-gptq-to-ggml.py Pavol Rusnak 2023-04-05 09:23:01 +0200
  • 57e9f929ee renamed misnamed ACCELERATE define, and removed all -march=native and -mtune=native flags Concedo 2023-04-05 15:22:13 +0800
  • 63cfa43200 quantize-stats: add option to test against reference quantization Håkon H. Hitland 2023-04-05 03:30:23 +0200
  • b1fa386c11
         Update ggml.c Sylvie 2023-04-04 21:50:15 +0200
  • dc679bf971
         Merge pull request #14 from hypnopump/update_macos Alexander 2023-04-04 21:42:45 +0500
  • d3801340f3
         streaming output hypnopump 2023-04-04 18:27:14 +0200
  • a9cb9adfd6
         streaming output hypnopump 2023-04-04 18:27:04 +0200
  • c320573b5e
         verify instructions can be followed hypnopump 2023-04-04 17:45:55 +0200
  • f5feb7470b
         verify instructions can be followed hypnopump 2023-04-04 17:45:06 +0200
  • b75a805563
         working on macos. no point in fp32 if all weights distributed in fp16 hypnopump 2023-04-04 17:39:21 +0200
  • 14273fea7a integrated gpt2 support Concedo 2023-04-04 23:15:47 +0800
  • 0349d03134 Add Accelerate framework dependency from Swift in this test. Andrew Duffy 2023-04-04 15:52:22 +0100
  • 42ad59fe41 Bugfix: We can handle the situation where matrix rows / thread count is not a multiple of TILESIZE_X Sebastian Apel 2023-04-04 16:23:51 +0200
  • 52de932842 removed main.exe to reduce clutter, added support for rep pen in gptj Concedo 2023-04-04 20:43:13 +0800
  • ce58bfc3ac Missing host optimizations in CXXFLAGS Ivan Stepanov 2023-04-04 15:33:01 +0300
  • 77e19980e9
         Merge pull request #13 from pixelkaiser/rwkv-macos Alexander 2023-04-04 14:24:21 +0500
  • 888db62c80
         Advise the kernel to preload the mapped memory Pavol Rusnak 2023-04-03 12:28:49 +0200
  • 5f1e91677c
         README: update with code-review for cmake build Adithya Balaji 2023-04-04 10:32:22 +0200
  • 977efba905 we actually build a dylib on macos PXLKSR 2023-04-04 10:19:06 +0200
  • a33cbbe03b Makefile: Added defaults for TILESIZE_X and _Y Sebastian Apel 2023-04-04 09:26:29 +0200
  • 9e4e917a96
         Merge branch 'ggerganov:master' into main barton ⊛ 2023-04-04 03:36:14 +0000
  • 890af8cacf
         Remove '[end_of_conversation]' line from Miku.sh at8u 2023-04-04 03:12:01 +0000
  • 6a4f137805 Fix wrongly copy-pasted mmap flags trollkotze 2023-04-04 03:19:09 +0200
  • 2c9910cd99 Change mmap parameters to avoid much swap thrashing trollkotze 2023-04-04 01:52:15 +0200
  • d4915074c4 quantize-stats: misc improvements Håkon H. Hitland 2023-04-04 00:33:09 +0200
  • 6d479decd7 Added support to compile MPI on Darwin Chad Brewbaker 2023-04-03 17:21:51 -0500
  • a7d3c3f304 quantize-stats: use less scratch memory Håkon H. Hitland 2023-04-04 00:21:01 +0200
  • 32d0fe7e92
         Trying again to fix error on windows compilation C2589: '(': illegal token CoderRC 2023-04-03 17:48:52 -0400
  • 634b09c9a4 add mpi Chad Brewbaker 2023-04-03 16:31:29 -0500
  • b90a3bf15e
         Trying to fix error on windows compilation C2589: '(': illegal token on right side of CoderRC 2023-04-03 17:13:25 -0400
  • 75eea96d01 Add benchmark script Sebastian Apel 2023-04-03 22:50:03 +0200
  • 8889c3be01
         Remove redundant duplicate #include <windows.h> CoderRC 2023-04-03 16:26:32 -0400
  • 10d758b917
         Change static pthread_create and pthread_join to non static pthread_create and pthread_join CoderRC 2023-04-03 16:08:42 -0400
  • 8c2ffe8559
         README: Update with CMake and windows example Adithya Balaji 2023-04-03 22:06:55 +0200
  • 0de310a159
         Remove deletions of Patch 2: Added threading for non posix systems CoderRC 2023-04-03 15:45:49 -0400
  • 361632264c Working version of tiled implementation Sebastian Apel 2023-04-03 21:20:55 +0200
  • 9881cb3301
         Improve ifdef logic Marco Matthies 2023-04-03 21:13:36 +0200
  • f43aca0f63 Simplify to include lower-case windows.h always, fix compilation on mingw32 Marco Matthies 2023-04-03 21:00:19 +0200
  • 8a7dd2c682
         Patch 2: Added threading for non posix systems CoderRC 2023-04-03 14:21:46 -0400
  • 68623ee175
         Merge branch 'ggerganov:master' into master CoderRC 2023-04-03 14:20:07 -0400
  • 37264707c2 Add "-e"/"--eval-threads" command-line parameter to set a different number of threads for single-token eval than for prompt eval. ml6 2023-04-03 11:17:07 -0700
  • 9c0dbbb08b Merge branch 'master' into concedo Concedo 2023-04-04 00:51:05 +0800
  • dd2abd8bc7 lower default thread threshold Concedo 2023-04-04 00:42:49 +0800
  • 53dbba7695
         Windows: reactive sigint handler after each Ctrl-C (#736) master-53dbba7 mgroeber9110 2023-04-03 18:00:55 +0200
  • 5b1143ed93 quantize-stats: show percentiles Håkon H. Hitland 2023-04-03 17:32:03 +0200
  • dc1c5ae7ec Experimental code that achives 30k FLOPS Sebastian Apel 2023-04-03 13:49:15 +0200
  • 437e77855a
         10+% performance improvement of ggml_vec_dot_q4_0 on AVX2 (#654) master-437e778 SebastianApel 2023-04-03 09:52:28 +0200
  • 1ed8878a4c Reviewer comments: removed double semicolon, deleted empty line 1962 Sebastian Apel 2023-04-03 09:31:15 +0200
  • 06c711d770 Merge branch 'master' into concedo Concedo 2023-04-03 15:10:08 +0800
  • aacc8b6872 Minor formatting changes saharNooby 2023-04-03 10:39:28 +0400
  • 4f1df7c89e
         Merge pull request #9 from hypnopump/more_instructions_works_linux Alexander 2023-04-03 11:35:38 +0500
  • fa74b016c6
         more details for macos/linux hypnopump 2023-04-03 08:33:57 +0200
  • b589e34f92 Fixed problem with MSVC compiler Sebastian Apel 2023-04-03 08:33:03 +0200
  • bea02c4b4c
         Merge branch 'master' into more_instructions_works_linux Eric Alcaide 2023-04-03 08:29:55 +0200
  • 0a0cabc4c7
         for consistency hypnopump 2023-04-03 08:27:00 +0200
  • 6f3fb01913
         suggestions hypnopump 2023-04-03 08:25:54 +0200
  • 3535476987 Update README.md: include info about pre-compiled library saharNooby 2023-04-03 09:48:53 +0400
  • 5b2830ed30 Increase memory for overhead from 32 MB to 256 MB saharNooby 2023-04-03 09:32:58 +0400
  • 61cd520cd6
         Patch 1: Added threading for non posix systems CoderRC 2023-04-03 00:53:39 -0400
  • ec59387899
         Added threading for non posix systems CoderRC 2023-04-03 00:41:39 -0400
  • 0a1c308d04
         Sync CoderRC 2023-04-02 23:52:27 -0400
  • 578c327dd4
         Fixed loading time by reading the file while letting the code execute CoderRC 2023-04-02 23:48:54 -0400
  • 864dcb26fb updates Gary Linscott 2023-04-02 20:16:15 -0700
  • eb5b22dda2 rebrand to koboldcpp Concedo 2023-04-03 10:35:18 +0800
  • c23078f57a
         Add --keep param to Miku.sh at8u 2023-04-03 02:15:33 +0000
  • bf18c6f4f9
         Update README.md Pi 2023-04-02 19:01:56 -0700
  • 0a5354fb1c
         Added cat translator Pi 2023-04-02 18:59:07 -0700
  • ed667e9581 quantize-stats command Håkon H. Hitland 2023-04-02 15:59:14 +0200
  • c79c99fb26
         CAT.CPP IS BETTER!!! Pi 2023-04-02 18:23:39 -0700
  • 3b419a52d8
         Made it talk like a cat Pi 2023-04-02 17:44:43 -0700
  • a17e745b6e Merge branch 'master' into batch_perplexity Gary Linscott 2023-04-02 17:44:30 -0700
  • cd7fa95690
         Define non-positive temperature behavior (#720) master-cd7fa95 Ivan Stepanov 2023-04-03 03:19:04 +0300
  • a64aaa81ec
         initial addition hypnopump 2023-04-03 00:52:26 +0200
  • 70420f0cac
         Add missing line to prompt in Miku.sh at8u 2023-04-02 22:36:07 +0000
  • a0c0516416
         Remove torch GPU dependencies from the Docker.full image (#665) bsilvereagle 2023-04-02 15:13:03 -0700
  • 9fe1524d02
         Remove pinning & use CPU whls directly bsilvereagle 2023-04-02 14:30:57 -0700
  • d0b53e82a3
         Add Miku.sh to examples at8u 2023-04-02 20:23:20 +0000
  • 69ef03d525 Performance improvement of AVX2 code Sebastian Apel 2023-04-02 22:13:08 +0200
  • 30ed065da1 Windows: reactive sigint handler after each Ctrl-C mgroeber9110 2023-04-02 22:04:44 +0200
  • 509c9460c6 Define non-positive temperature behavior Ivan Stepanov 2023-04-02 23:02:23 +0300
  • 56c78d16d2 Reviewer feedback: Updated clean: and benchmark: sections Sebastian Apel 2023-04-02 21:16:43 +0200
  • 8dd8ab1659 Various enhancement and integration pygmalion.cpp Concedo 2023-04-03 00:04:43 +0800
  • d62a050144 Remove hardcoded memory requirements table saharNooby 2023-04-02 18:37:45 +0400
  • b31447507a Update gptq to ggml loader as well albanD 2023-04-02 09:32:27 -0400
  • 1262ad0456 Fix build errors and warnings saharNooby 2023-04-02 17:23:39 +0400
  • f2b1dad22b Add GitHub workflows file saharNooby 2023-04-02 16:56:04 +0400