]>
git.djapps.eu Git - pkg/ggml/sources/llama.cpp/log
Qingyou Meng [Mon, 20 Mar 2023 19:33:10 +0000 (03:33 +0800)]
Fixed tokenizer.model not found error when model dir is symlink (#325)
Mack Straight [Mon, 20 Mar 2023 19:26:01 +0000 (12:26 -0700)]
move file magic/version to header, print expected version (#319)
Bernat Vadell [Mon, 20 Mar 2023 17:05:20 +0000 (18:05 +0100)]
Docker - Fix publish docker image in GitHub Registry (#235)
* fix publish permission
* try to fix docker pipeline using as password github_token & username repository_owner
Mack Straight [Mon, 20 Mar 2023 10:17:23 +0000 (03:17 -0700)]
sentencepiece bpe compatible tokenizer (#252)
* potential out of bounds read
* fix quantize
* style
* Update convert-pth-to-ggml.py
* mild cleanup
* don't need the space-prefixing here rn since main.cpp already does it
* new file magic + version header field
* readme notice
* missing newlines
Co-authored-by: slaren <redacted>
Stephan Walter [Mon, 20 Mar 2023 08:24:11 +0000 (08:24 +0000)]
Add tqdm to Python requirements (#293)
* Add tqdm to Python requirements
* Remove torchvision torchaudio, add requests
cocktailpeanut [Sun, 19 Mar 2023 21:44:20 +0000 (17:44 -0400)]
bugfix: default should not be interactive (#304)
Georgi Gerganov [Sun, 19 Mar 2023 19:58:51 +0000 (21:58 +0200)]
Rename script
Georgi Gerganov [Sun, 19 Mar 2023 19:57:28 +0000 (21:57 +0200)]
Add temporary helper script for Alpaca chat
Rickey Bowers Jr [Sun, 19 Mar 2023 19:44:30 +0000 (13:44 -0600)]
fix coloring of last `n_batch` of prompt, and refactor line input (#221)
* fix coloring of last `n_batch` of prompt, and refactor line input
* forgot the newline that needs to be sent to the model
* (per #283) try to force flush of color reset in SIGINT handler
tjohnman [Sun, 19 Mar 2023 19:33:06 +0000 (20:33 +0100)]
Support for multiple reverse prompts. (#299)
Co-authored-by: Johnman <>
Co-authored-by: Johnman <redacted>
Suaj Carrot [Sun, 19 Mar 2023 18:38:44 +0000 (12:38 -0600)]
Improved quantize script (#222)
* Improved quantize script
I improved the quantize script by adding error handling and allowing to select many models for quantization at once in the command line. I also converted it to Python for generalization as well as extensibility.
* Fixes and improvements based on Matt's observations
Fixed and improved many things in the script based on the reviews made by @mattsta. The parallelization suggestion is still to be revised, but code for it was still added (commented).
* Small fixes to the previous commit
* Corrected to use the original glob pattern
The original Bash script uses a glob pattern to match files that have endings such as ...bin.0, ...bin.1, etc. That has been translated correctly to Python now.
* Added support for Windows and updated README to use this script
New code to set the name of the quantize script binary depending on the platform has been added (quantize.exe if working on Windows) and the README.md file has been updated to use this script instead of the Bash one.
* Fixed a typo and removed shell=True in the subprocess.run call
Fixed a typo regarding the new filenames of the quantized models and removed the shell=True parameter in the subprocess.run call as it was conflicting with the list of parameters.
* Corrected previous commit
* Small tweak: changed the name of the program in argparse
This was making the automatic help message to be suggesting the program's usage as being literally "$ Quantization Script [arguments]". It should now be something like "$ python3 quantize.py [arguments]".
tjohnman [Sun, 19 Mar 2023 18:36:19 +0000 (19:36 +0100)]
Make prompt randomization optional. (#300)
Co-authored-by: Johnman <>
tjohnman [Sun, 19 Mar 2023 18:31:17 +0000 (19:31 +0100)]
Respect the maximum number of tokens in interactive. (#298)
Co-authored-by: Johnman <redacted>
Co-authored-by: Georgi Gerganov <redacted>
slaren [Sun, 19 Mar 2023 18:22:48 +0000 (19:22 +0100)]
Add --ignore-eos parameter (#181)
Co-authored-by: Georgi Gerganov <redacted>
Qingyou Meng [Sun, 19 Mar 2023 18:10:00 +0000 (02:10 +0800)]
interactive mode: print '\n' in sigint_handler, this flush stdout thus ensure color reset. (#283)
Erik Scholz [Sun, 19 Mar 2023 17:57:00 +0000 (18:57 +0100)]
Command line switch to use F16 for memory_k and memory_v (refactor of #154) (#294)
* Use F16 for memory_k and memory_v
* add command line switch to use f16 instead of f32 for memory k+v
---------
Co-authored-by: Ty Everett <redacted>
Georgi Gerganov [Sun, 19 Mar 2023 17:51:55 +0000 (19:51 +0200)]
Update hot topics to mention Alpaca support
Georgi Gerganov [Sun, 19 Mar 2023 17:46:32 +0000 (19:46 +0200)]
Fix off-by-one bug (#115)
Georgi Gerganov [Sun, 19 Mar 2023 17:33:18 +0000 (19:33 +0200)]
Fix python stuff (#109)
qunash [Sun, 19 Mar 2023 17:17:39 +0000 (20:17 +0300)]
Refactoring `convert-pth-to-ggml.py`: more concise and readable (#109)
* Refactor get_n_parts function to simplify code and improve readability
* Use f-strings instead of concatenation
* Refactoring: more concise and readable
* modularize
---------
Co-authored-by: Georgi Gerganov <redacted>
Georgi Gerganov [Sun, 19 Mar 2023 17:04:44 +0000 (19:04 +0200)]
Drop trailing new line from file prompts (#80)
Georgi Gerganov [Sun, 19 Mar 2023 16:49:50 +0000 (18:49 +0200)]
Add instruction for using Alpaca (#240)
Georgi Gerganov [Sun, 19 Mar 2023 16:37:02 +0000 (18:37 +0200)]
Add "--instruct" argument for usage with Alpaca (#240)
Also start adding prompts in "./prompts"
Georgi Gerganov [Sun, 19 Mar 2023 15:30:00 +0000 (17:30 +0200)]
Change RMSNorm eps to 1e-6 (#173)
I think this is what is used in the Python code
Ronsor [Sun, 19 Mar 2023 00:10:47 +0000 (17:10 -0700)]
Warn user if a context size greater than 2048 tokens is specified (#274)
LLaMA doesn't support more than 2048 token context sizes, and going above that produces terrible results.
Pavol Rusnak [Sat, 18 Mar 2023 21:39:46 +0000 (22:39 +0100)]
Fix typo in readme
Pavol Rusnak [Sat, 18 Mar 2023 21:20:04 +0000 (22:20 +0100)]
Add note about Python 3.11 to readme
Pavol Rusnak [Sat, 18 Mar 2023 20:58:46 +0000 (21:58 +0100)]
Add memory/disk requirements to readme
Alex Nguyen [Sat, 18 Mar 2023 13:51:49 +0000 (20:51 +0700)]
Remove unused code since n_vocab is model.hparams.n_vocab (#262)
Justin Suess [Sat, 18 Mar 2023 11:44:09 +0000 (07:44 -0400)]
fixed warning with std::ignore about unused function result (#151)
fixed warning with std::ignore about unused function result
Gary Linscott [Sat, 18 Mar 2023 11:17:19 +0000 (04:17 -0700)]
Fix n^2 loop in tokenization (#254)
This causes long prompts to parse very slowly.
anzz1 [Sat, 18 Mar 2023 07:27:12 +0000 (09:27 +0200)]
CI Improvements (#230)
* CI Improvements
Manual build feature, autoreleases for Windows
* better CI naming convention
use branch name in releases and tags
Niklas Korz [Fri, 17 Mar 2023 22:03:48 +0000 (23:03 +0100)]
Nix flake (#40)
* Nix flake
* Nix: only add Accelerate framework on macOS
* Nix: development shel, direnv and compatibility
* Nix: use python packages supplied by withPackages
* Nix: remove channel compatibility
* Nix: fix ARM neon dotproduct on macOS
---------
Co-authored-by: Pavol Rusnak <redacted>
thement [Fri, 17 Mar 2023 20:05:58 +0000 (21:05 +0100)]
Implement non-greedy tokenizer that tries to maximize token lengths (#242)
* Implement non-greedy tokenizer that tries to maximize token lengths
* Insert single space in front of the prompt
- this is to match original llama tokenizer behavior
---------
Co-authored-by: Jakub Horak <redacted>
Georgi Gerganov [Fri, 17 Mar 2023 19:46:46 +0000 (21:46 +0200)]
Default to 4 threads (#243)
Georgi Gerganov [Fri, 17 Mar 2023 18:30:04 +0000 (20:30 +0200)]
Update Contributing section
Stephan Walter [Fri, 17 Mar 2023 17:47:35 +0000 (17:47 +0000)]
Don't tell users to use a bad number of threads (#243)
The readme tells people to use the command line option "-t 8", causing 8
threads to be started. On systems with fewer than 8 cores, this causes a
significant slowdown. Remove the option from the example command lines
and use /proc/cpuinfo on Linux to determine a sensible default.
mmyjona [Fri, 17 Mar 2023 16:38:24 +0000 (00:38 +0800)]
add ptread link to fix cmake build under linux (#114)
* add ptread link to fix cmake build under linux
* add cmake to linux and macos platform
* separate make and cmake workflow
---------
Co-authored-by: SebastiƔn A <redacted>
Bernat Vadell [Fri, 17 Mar 2023 09:47:06 +0000 (10:47 +0100)]
š Dockerize llamacpp (#132)
* feat: dockerize llamacpp
* feat: split build & runtime stages
* split dockerfile into main & tools
* add quantize into tool docker image
* Update .devops/tools.sh
Co-authored-by: Georgi Gerganov <redacted>
* add docker action pipeline
* change CI to publish at github docker registry
* fix name runs-on macOS-latest is macos-latest (lowercase)
* include docker versioned images
* fix github action docker
* fix docker.yml
* feat: include all-in-one command tool & update readme.md
---------
Co-authored-by: Georgi Gerganov <redacted>
Matvey Soloviev [Fri, 17 Mar 2023 04:48:39 +0000 (05:48 +0100)]
Q4_1 quantization (#193)
* Add AVX2 version of ggml_vec_dot_q4_1
* Small optimisations to q4_1 dot product (@Const-me)
* Rearrange Q4_1 quantization to work for multipart models. (Fix #152)
* Fix ggml_vec_mad_q4_1 too
* Fix non-vectorised q4_1 vec mul
Georgi Gerganov [Thu, 16 Mar 2023 13:00:09 +0000 (15:00 +0200)]
Update README.md
Georgi Gerganov [Thu, 16 Mar 2023 06:55:13 +0000 (08:55 +0200)]
Expand "Contributing" section
Georgi Gerganov [Thu, 16 Mar 2023 05:12:12 +0000 (07:12 +0200)]
Update hot topics - RMSnorm
Nebula [Wed, 15 Mar 2023 23:29:25 +0000 (19:29 -0400)]
Fix RMS norm in GGML (#191)
hoangmit [Wed, 15 Mar 2023 22:41:38 +0000 (18:41 -0400)]
Add RMS norm and use it (#187)
* add ggml_rms_norm
* update op num
moritzbrantner [Wed, 15 Mar 2023 20:35:25 +0000 (21:35 +0100)]
fixed typo (#178)
Rickey Bowers Jr [Wed, 15 Mar 2023 19:56:24 +0000 (13:56 -0600)]
add SIGINT support for _WIN32 environments (#120)
* add SIGINT support for _WIN32 environments
* perhaps more consistent
Justin Suess [Wed, 15 Mar 2023 19:42:40 +0000 (15:42 -0400)]
added ctx_size parameter (#148)
* added ctx_size parameter
* added it in more places
* Apply suggestions from code review
---------
Co-authored-by: Georgi Gerganov <redacted>
Justin Suess [Wed, 15 Mar 2023 19:39:38 +0000 (15:39 -0400)]
fixed color reset on exit (#149)
* fixed color reset on exit
* added sigint handler for ansi_color_reset
* Update main.cpp
---------
Co-authored-by: Georgi Gerganov <redacted>
Musab Gultekin [Wed, 15 Mar 2023 19:39:06 +0000 (22:39 +0300)]
Fix potential licensing issue (#126)
* Update README.md
* Update README.md
remove facebook
Ronsor [Wed, 15 Mar 2023 19:37:50 +0000 (12:37 -0700)]
Use `tokenizer.vocab_size()` instead of hardcoding 32000 in convert-pth-to-ggml.py (#142)
There are ways that special tokens or other new tokens could be added to the tokenizer; therefore it's probably best not to assume the vocabulary is only 32000 tokens.
hoangmit [Wed, 15 Mar 2023 19:05:14 +0000 (15:05 -0400)]
inline -> static inline for "bytesFromNibbles" (#161)
Without "static" prefix, it fails to compile in clang
Ronsor [Tue, 14 Mar 2023 19:34:37 +0000 (12:34 -0700)]
Don't use vdotq_s32 if it's not available (#139)
* Don't use vdotq_s32 if it's not available
`dotprod` extensions aren't available on some ARM CPUs (e.g. Raspberry Pi 4), so check for them and only use them if they're available.
Reintroduces the code removed in
84d9015 if `__ARM_FEATURE_DOTPROD` isn't defined.
* Update ggml.c
---------
Co-authored-by: Georgi Gerganov <redacted>
Radoslav Gerganov [Tue, 14 Mar 2023 13:30:08 +0000 (15:30 +0200)]
Add section to README on how to run the project on Android (#130)
Georgi Gerganov [Tue, 14 Mar 2023 07:43:52 +0000 (09:43 +0200)]
Add Misc section + update hot topics + minor fixes
SebastiƔn A [Mon, 13 Mar 2023 20:29:10 +0000 (17:29 -0300)]
Add windows to the CI (#98)
Georgi Gerganov [Mon, 13 Mar 2023 19:22:15 +0000 (21:22 +0200)]
CMake build in Release by default (#75)
Georgi Gerganov [Mon, 13 Mar 2023 17:21:51 +0000 (19:21 +0200)]
Update contribution section, hot topics, limitations, etc.
Georgi Gerganov [Mon, 13 Mar 2023 17:15:08 +0000 (19:15 +0200)]
Print system information
SebastiƔn A [Mon, 13 Mar 2023 17:12:33 +0000 (14:12 -0300)]
Initial support for CMake (#75)
Thomas Klausner [Mon, 13 Mar 2023 16:40:54 +0000 (17:40 +0100)]
Add NetBSD support. (#90)
Pavol Rusnak [Mon, 13 Mar 2023 16:39:56 +0000 (17:39 +0100)]
Use fprintf for diagnostic output (#48)
keep printf only for printing model output
one can now use ./main ... 2>dev/null to suppress any diagnostic output
Georgi Gerganov [Mon, 13 Mar 2023 16:36:44 +0000 (18:36 +0200)]
Use vdotq_s32 to improve performance (#67)
* 10% performance boost on ARM
* Back to original change
uint256_t [Mon, 13 Mar 2023 16:33:43 +0000 (01:33 +0900)]
Reduce model loading time (#43)
* Use buffering
* Use vector
* Minor
---------
Co-authored-by: Georgi Gerganov <redacted>
Val Kharitonov [Mon, 13 Mar 2023 16:24:18 +0000 (12:24 -0400)]
Fix UTF-8 handling (including colors) (#79)
Pavol Rusnak [Mon, 13 Mar 2023 16:15:20 +0000 (17:15 +0100)]
Add quantize script for batch quantization (#92)
* Add quantize script for batch quantization
* Indentation
* README for new quantize.sh
* Fix script name
* Fix file list on Mac OS
---------
Co-authored-by: Georgi Gerganov <redacted>
Georgi Gerganov [Mon, 13 Mar 2023 07:42:26 +0000 (09:42 +0200)]
Add initial contribution guidelines
Matvey Soloviev [Mon, 13 Mar 2023 03:08:01 +0000 (04:08 +0100)]
Gate signal support on being on a unixoid system. (#74)
Matvey Soloviev [Sun, 12 Mar 2023 23:35:51 +0000 (00:35 +0100)]
Fix token count accounting
Georgi Gerganov [Sun, 12 Mar 2023 23:28:08 +0000 (01:28 +0200)]
Revert "10% performance boost on ARM"
This reverts commit
113a9e83ebc0f788f861394437087bf3ca0e019b .
There are some reports for illegal instruction.
Moved this stuff to vdotq_s32 branch until resolve
Georgi Gerganov [Sun, 12 Mar 2023 23:21:03 +0000 (01:21 +0200)]
Check for vdotq_s32 availability
Georgi Gerganov [Sun, 12 Mar 2023 23:05:24 +0000 (01:05 +0200)]
Ammend to previous commit - forgot to update non-QRDMX branch
Georgi Gerganov [Sun, 12 Mar 2023 22:56:10 +0000 (00:56 +0200)]
10% performance boost on ARM
Matvey Soloviev [Sun, 12 Mar 2023 22:07:34 +0000 (23:07 +0100)]
Fix color getting reset before prompt output done (#65)
(cherry picked from commit
7eb2987619feee04c40eff69b604017d09919cb6 )
Georgi Gerganov [Sun, 12 Mar 2023 21:39:01 +0000 (23:39 +0200)]
Update README.md
Matvey Soloviev [Sun, 12 Mar 2023 21:13:28 +0000 (22:13 +0100)]
Add interactive mode (#61)
* Initial work on interactive mode.
* Improve interactive mode. Make rev. prompt optional.
* Update README to explain interactive mode.
* Fix OS X build
Marc Kƶhlbrugge [Sun, 12 Mar 2023 20:30:08 +0000 (03:30 +0700)]
Fix typo in README (#45)
Ben Garney [Sun, 12 Mar 2023 20:28:36 +0000 (13:28 -0700)]
Allow using prompt files (#59)
beiller [Sun, 12 Mar 2023 20:23:15 +0000 (16:23 -0400)]
Add back top_k (#56)
* Add back top_k
* Update utils.cpp
* Update utils.h
---------
Co-authored-by: Bill Hamilton <redacted>
Co-authored-by: Georgi Gerganov <redacted>
SebastiƔn A [Sun, 12 Mar 2023 20:15:00 +0000 (17:15 -0300)]
Windows fixes (#31)
* Apply fixes suggested to build on windows
Issue: https://github.com/ggerganov/llama.cpp/issues/22
* Remove unsupported VLAs
* MSVC: Remove features that are only available on MSVC C++20.
* Fix zero initialization of the other fields.
* Change the use of vector for stack allocations.
Georgi Gerganov [Sun, 12 Mar 2023 20:09:26 +0000 (22:09 +0200)]
Update README.md
Georgi Gerganov [Sun, 12 Mar 2023 20:08:24 +0000 (22:08 +0200)]
Add CI (#60)
Georgi Gerganov [Sun, 12 Mar 2023 18:59:01 +0000 (20:59 +0200)]
Revert "weights_only" arg - this causing more trouble than help
Oleksandr Nikitin [Sun, 12 Mar 2023 12:16:33 +0000 (14:16 +0200)]
python/pytorch compat notes (#44)
beiller [Sun, 12 Mar 2023 09:27:42 +0000 (05:27 -0400)]
Add repetition penalty (#20)
* Adding repeat penalization
* Update utils.h
* Update utils.cpp
* Numeric fix
Should probably still scale by temp even if penalized
* Update comments, more proper application
I see that numbers can go negative so a fix from a referenced commit
* Minor formatting
---------
Co-authored-by: Georgi Gerganov <redacted>
Georgi Gerganov [Sun, 12 Mar 2023 07:03:25 +0000 (09:03 +0200)]
Clarify meaning of hacking
Georgi Gerganov [Sun, 12 Mar 2023 06:41:54 +0000 (08:41 +0200)]
README: add "Supported platforms" + update hot topics
deepdiffuser [Sun, 12 Mar 2023 06:36:35 +0000 (22:36 -0800)]
use weights_only in conversion script (#32)
this restricts malicious weights from executing arbitrary code by restricting the unpickler to only loading tensors, primitive types, and dictionaries
Pavol Rusnak [Sun, 12 Mar 2023 06:36:03 +0000 (07:36 +0100)]
Add LICENSE (#21)
Georgi Gerganov [Sat, 11 Mar 2023 23:26:32 +0000 (01:26 +0200)]
Update README.md
Juraj Bednar [Sat, 11 Mar 2023 17:32:20 +0000 (18:32 +0100)]
Fix a typo in model name (#16)
Georgi Gerganov [Sat, 11 Mar 2023 16:10:18 +0000 (18:10 +0200)]
Update README.md
Georgi Gerganov [Sat, 11 Mar 2023 15:58:18 +0000 (17:58 +0200)]
Add AVX2 support for x86 architectures thanks to @Const-me !
Georgi Gerganov [Sat, 11 Mar 2023 15:40:14 +0000 (17:40 +0200)]
Fix un-initialized FP16 tables on x86 (#15, #2)
Georgi Gerganov [Sat, 11 Mar 2023 10:44:21 +0000 (12:44 +0200)]
Bump memory buffer
Georgi Gerganov [Sat, 11 Mar 2023 10:31:21 +0000 (12:31 +0200)]
Update README.md
Georgi Gerganov [Sat, 11 Mar 2023 10:26:46 +0000 (12:26 +0200)]
.gitignore models/
Georgi Gerganov [Sat, 11 Mar 2023 10:26:16 +0000 (12:26 +0200)]
Update Makefile var + add comment
Georgi Gerganov [Sat, 11 Mar 2023 09:34:25 +0000 (11:34 +0200)]
Update README.md
Georgi Gerganov [Sat, 11 Mar 2023 09:34:11 +0000 (11:34 +0200)]
Update README.md