Commit Graph

1410 Commits

Author SHA1 Message Date
Georgi Gerganov
684da25926
ggml : fix quantize_row_q4_1() ARM_NEON (close #876) 2023-04-10 19:29:48 +03:00
comex
180b693a47 Print model version.
Also improve model type printing, and fix indentation of an unrelated
switch statement.
2023-04-10 01:10:46 +02:00
comex
f963b63afa Rewrite loading code to try to satisfy everyone:
- Support all three formats (ggml, ggmf, ggjt).  (However, I didn't
  include the hack needed to support GPT4All files without conversion.
  Those can still be used after converting them with convert.py from my
  other PR.)

- Support both mmap and read (mmap is used by default, but can be
  disabled with `--no-mmap`, and is automatically disabled for pre-ggjt
  files or on platforms where mmap is not supported).

- Support multi-file models like before, but automatically determine the
  number of parts rather than requiring `--n_parts`.

- Improve validation and error checking.

- Stop using the per-file type field (f16) entirely in favor of just
  relying on the per-tensor type/size fields.  This has no immediate
  benefit, but makes it easier to experiment with different formats, and
  should make it easier to support the new GPTQ-for-LLaMa models in the
  future (I have some work in progress on that front).

- Support VirtualLock on Windows (using the same `--mlock` option as on
  Unix).

    - Indicate loading progress when using mmap + mlock.  (Which led me
      to the interesting observation that on my Linux machine, with a
      warm file cache, mlock actually takes some time, whereas mmap
      without mlock starts almost instantly...)

      - To help implement this, move mlock support from ggml to the
        loading code.

- madvise/PrefetchVirtualMemory support (based on #740)

- Switch from ifstream to the `fopen` family of functions to avoid
  unnecessary copying and, when mmap is enabled, allow reusing the same
  file descriptor for both metadata reads and mmap (whereas the existing
  implementation opens the file a second time to mmap).

- Quantization now produces a single-file output even with multi-file
  inputs (not really a feature as much as 'it was easier this way').

Implementation notes:

I tried to factor the code into more discrete pieces than before.

Regarding code style: I tried to follow the code style, but I'm naughty
and used a few advanced C++ features repeatedly:

- Destructors to make it easier to ensure everything gets cleaned up.

- Exceptions.  I don't even usually use exceptions when writing C++, and
  I can remove them if desired... but here they make the loading code
  much more succinct while still properly handling a variety of errors,
  ranging from API calls failing to integer overflow and allocation
  failure.  The exceptions are converted to error codes at the
  API boundary.)

Co-authored-by: Pavol Rusnak <pavol@rusnak.io> (for the bit I copied from #740)
2023-04-10 01:10:46 +02:00
Tomáš Pazdiora
aaf3b23deb
fix for windows utf-8 input (#840)
Use UTF-16 as input on Windows, since UTF-8 does not work and reads multibyte characters as zeros
2023-04-08 17:49:39 +02:00
eiery
f2d1c47294
cmake should link openblas properly with -lopenblas like how it's done in the makefile (#839) 2023-04-08 11:15:17 +00:00
lon
317fb12fbd
Add new binaries to flake.nix (#847) 2023-04-08 12:04:23 +02:00
unbounded
62cfc54f77
Add quantize-stats command for testing quantization (#728)
Command that calculates some statistics over the errors introduced by
quantization, like mean square error, max error and some percentile errors for layer
weights. Should be useful for testing quantization improvements.

Exposes some internal state from ggml and llama for testing
2023-04-08 00:09:18 +02:00
bhubbb
698f7b5d63
make : add libllama.so target for llama-cpp-python (#797)
I was able to get llama-cpp-python working but only when I build libllama.so with make.
2023-04-07 19:11:58 +03:00
iacore
c1950c3431
zig : don't link examples/common.cpp for non-example (#814) 2023-04-07 19:05:29 +03:00
Ivan Stepanov
4953e9007f
llama : always sort logits before nucleus sampling (#812)
* Always sort logits before nucleus sampling

* remove second normalization

- fix windows build
- remove normalization since std::discrete_distribution does not require it
2023-04-07 19:02:12 +03:00
Sergey Alirzaev
cc9cee8e9e
Do not crash when it has nothing to say. (#796)
Otherwise observing this in the interactive mode:
/usr/lib/gcc/x86_64-pc-linux-gnu/12/include/g++-v12/bits/stl_vector.h:1230: reference std::vector<int>::back() [_Tp = int, _Alloc = std::allocator<int>]: Assertion '!this->empty()' failed.
2023-04-06 17:59:11 +02:00
Pavol Rusnak
d2beca95dc
Make docker instructions more explicit (#785) 2023-04-06 08:56:58 +02:00
Georgi Gerganov
eeaa7b0492
ggml : multi-thread ggml_rope() (~3-4 times faster on M1) (#781) 2023-04-05 22:11:03 +03:00
Georgi Gerganov
986b6ce9f9
ggml, llama : avoid heavy V transpose + improvements (#775)
ggml :

- added ggml_view_3d()
- ggml_view_tensor() now inherits the stride too
- reimplement ggml_cpy() to account for dst stride
- no longer require tensor->data to be memory aligned

llama :

- compute RoPE on 32-bit tensors (should be more accurate)
- store RoPE-ed K in the KV cache
- store transposed V in the KV cache (significant speed-up)
- avoid unnecessary Q copy
2023-04-05 22:07:33 +03:00
Georgi Gerganov
3416298929
Update README.md 2023-04-05 19:54:30 +03:00
Ivan Stepanov
5a8c4f6240
llama : define non-positive top_k; top_k range check (#779)
* Define non-positive top_k; top_k range check

* minor : brackets

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2023-04-05 19:20:05 +03:00
at8u
ff05d05c96
miku.sh : add executable bit (#780) 2023-04-05 18:59:13 +03:00
Georgi Gerganov
62b3e81aae
media : add logos and banners 2023-04-05 18:58:31 +03:00
Georgi Gerganov
8d10406d6e
readme : change logo + add bindings + add uis + add wiki 2023-04-05 18:56:20 +03:00
iacore
ed1c214e66
zig : add build.zig (#773)
Co-authored-by: Locria Cyber <74560659+locriacyber@users.noreply.github.com>
2023-04-05 18:06:02 +03:00
Ivan Stepanov
0c44427df1
make : missing host optimizations in CXXFLAGS (#763) 2023-04-05 17:38:37 +03:00
Adithya Balaji
594cc95fab
readme : update with CMake and windows example (#748)
* README: Update with CMake and windows example

* README: update with code-review for cmake build
2023-04-05 17:36:12 +03:00
at8u
88ed5761b8
examples : add Miku.sh (#724)
* Add Miku.sh to examples

* Add missing line to prompt in Miku.sh

* Add --keep param to Miku.sh

* Remove '[end_of_conversation]' line from Miku.sh

No longer is necessary.
2023-04-05 17:32:42 +03:00
Andrew Duffy
58c438cf7d
Add Accelerate/BLAS when using Swift (#765) 2023-04-05 06:44:24 -04:00
mgroeber9110
53dbba7695
Windows: reactive sigint handler after each Ctrl-C (#736) 2023-04-03 18:00:55 +02:00
SebastianApel
437e77855a
10+% performance improvement of ggml_vec_dot_q4_0 on AVX2 (#654)
* Performance improvement of AVX2 code
* Fixed problem with MSVC compiler
* Reviewer comments: removed double semicolon, deleted empty line 1962
2023-04-03 09:52:28 +02:00
Ivan Stepanov
cd7fa95690
Define non-positive temperature behavior (#720) 2023-04-03 02:19:04 +02:00
bsilvereagle
a0c0516416
Remove torch GPU dependencies from the Docker.full image (#665)
By using `pip install torch --index-url https://download.pytorch.org/whl/cpu`
instead of `pip install torch` we can specify we want to install a CPU-only version
of PyTorch without any GPU dependencies. This reduces the size of the Docker image
from 7.32 GB to 1.62 GB
2023-04-03 00:13:03 +02:00
Thatcher Chamberlin
d8d4e865cd
Add a missing step to the gpt4all instructions (#690)
`migrate-ggml-2023-03-30-pr613.py` is needed to get gpt4all running.
2023-04-02 12:48:57 +02:00
Christian Falch
e986f94829
Added api for getting/setting the kv_cache (#685)
The api provides access methods for retrieving the current memory buffer for the kv_cache and its token number.
It also contains a method for setting the kv_cache from a memory buffer.

This makes it possible to load/save history - maybe support --cache-prompt paramater as well?

Co-authored-by: Pavol Rusnak <pavol@rusnak.io>
2023-04-02 12:23:04 +02:00
Marian Cepok
c0bb1d3ce2
ggml : change ne to int64_t (#626) 2023-04-02 13:21:31 +03:00
Leonardo Neumann
6e7801d08d
examples : add gpt4all script (#658) 2023-04-02 10:56:20 +03:00
Stephan Walter
81040f10aa
llama : do not allocate KV cache for "vocab_only == true" (#682)
Fixes sanitizer CI
2023-04-02 10:18:53 +03:00
Fabian
c4f89d8d73
make : use -march=native -mtune=native on x86 (#609) 2023-04-02 10:17:05 +03:00
Murilo Santana
5b70e7de4c
fix default params for examples/main (#697) 2023-04-02 04:41:12 +02:00
Ikko Eltociear Ashimine
a717cba844
py: huggingface -> Hugging Face (#686) 2023-04-01 18:38:18 +02:00
rimoliga
d0a7f742e7
readme: replace termux links with homepage, play store is deprecated (#680) 2023-04-01 16:57:30 +02:00
Slaren
0d054e292e Show error message when -f fails 2023-04-01 16:08:40 +02:00
Stephan Walter
3525899277
Enable -std= for cmake builds, fix warnings (#598) 2023-03-31 19:19:16 +00:00
slaren
1d08882afa
Optimize AVX2 ggml_vec_dot_q4_0 (#642) 2023-03-31 15:55:52 +00:00
perserk
02c5b27e91
Add AVX acceleration (#617)
* ggml : add AVX quantize_row_q4_0()

* ggml : add AVX ggml_vec_dot_q4_0()

* ggml : refactor AVX part of ggml_vec_dot_q4_0()

https://github.com/ggerganov/llama.cpp/pull/617#issuecomment-1489985645
2023-03-31 13:55:44 +02:00
Pavol Rusnak
cbef542879 py : cleanup the code
- use f-strings where possible
- drop first param of encode/decode functions since "utf-8" is the default
2023-03-31 10:32:01 +02:00
Pavol Rusnak
9733104be5 drop quantize.py (now that models are using a single file) 2023-03-31 01:07:32 +02:00
Georgi Gerganov
3df890aef4
readme : update supported models 2023-03-30 22:31:54 +03:00
Justine Tunney
ee0c40dd6d Introduce GGML migration tool for new file format
If you deleted your old Meta LLaMA .pth files, then the
migrate-ggml-2023-03-30-pr613.py script will allow you to convert your
old ggml files into the new mmap()'able format.

See #613
2023-03-30 12:28:25 -07:00
Justine Tunney
6f23ba5ee2 Ensure --mlock works properly with mmap() support 2023-03-30 12:28:25 -07:00
Justine Tunney
78ca9838ee Make loading weights 10-100x faster
This is a breaking change that's going to give you three benefits:

1. Your inference commands should load 100x faster
2. You may be able to safely load models 2x larger
3. You can run many concurrent inference processes

This was accomplished by changing the file format so we can mmap()
weights directly into memory without having to read() or copy them
thereby ensuring the kernel can make its file cache pages directly
accessible to our inference processes; and secondly, that the file
cache pages are much less likely to get evicted (which would force
loads to hit disk) because they're no longer competing with memory
pages that were needlessly created by gigabytes of standard i/o.

The new file format supports single-file models like LLaMA 7b, and
it also supports multi-file models like LLaMA 13B. Our Python tool
now merges the foo.1, foo.2, etc. files back into a single file so
that the C++ code which maps it doesn't need to reshape data every
time. That's made llama.cpp so much simpler. Much of its load code
has now been deleted.

Furthermore, this change ensures that tensors are aligned properly
on a 32-byte boundary. That opens the door to seeing if we can get
additional performance gains on some microprocessors, by using ops
that require memory alignment.

Lastly note that both POSIX and the Windows platform are supported

Fixes #91
2023-03-30 12:28:25 -07:00
Slaren
a017390358 Initial windows support (untested) 2023-03-30 12:28:25 -07:00
Slaren
ac184d5147 Always initialize mm_addr and mm_length in llama_model 2023-03-30 12:28:25 -07:00
Slaren
276e5b7811 Unmap the file in llama_free 2023-03-30 12:28:25 -07:00