Commit Graph

929 Commits

Author SHA1 Message Date
161b51d91a CUDA: faster dequantize kernels for Q4_0 and Q4_1 (llama/4938)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-17 21:21:09 +02:00
f904b31a7d Add ability to use importance matrix for all k-quants (llama/4930)
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-17 21:21:09 +02:00
f6614155e4 talk-llama : optional wake-up command and audio confirmation (#1765)
* talk-llama: add optional wake-word detection from command

* talk-llama: add optional audio confirmation before generating answer

* talk-llama: fix small formatting issue in output

* talk-llama.cpp: fix Windows build
2024-01-16 15:52:01 +02:00
f5f159c320 server : fix building and simplify lib deps on Windows (#1772)
* make : fix server example building on MSYS2 environments (Windows)

It was not working since commit eff3570f78
when server was introduced.

* cmake : simplify server example lib deps on Windows

server uses httplib::Server, not httplib::SSLServer, so there is no need
to mention cryptographic libraries in target_link_libraries.
Winsock (ws2_32) suffices here.

Also use plain library names like we use in other places.
2024-01-15 15:48:13 +02:00
6ebba525f1 talk-llama : sync llama.cpp 2024-01-14 18:08:20 +02:00
2a5874441d talk-llama : llama.cpp 2024-01-14 11:06:28 +02:00
d08445c9ad sync : ggml 2024-01-14 10:55:18 +02:00
4a945696cb metal : correctly set SIMD support flags on iOS (llama/4923)
* Correctly set support_simdgroup_reduction and support_simdgroup_mm on iPhone/iPad

* log a little bit more info on iOS
2024-01-14 10:54:09 +02:00
dabc964d83 2-bit quantizations (llama/4897)
* imatrix: load

* imatrix: WIP

* imatrix: Add Q2_K quantization

* imatrix: also guard against Q2_K_S quantization without importance matrix

* imatrix: guard even more against low-bit quantization misuse

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-14 10:54:09 +02:00
654baf693d scripts : sync-ggml-am.sh add option to skip commits 2024-01-14 10:53:19 +02:00
f001a3b7b6 talk-llama : sync llama.cpp 2024-01-14 00:13:17 +02:00
c615f2c335 sync : ggml 2024-01-14 00:12:17 +02:00
d839dd0242 examples : adapt to metal API 2024-01-14 00:11:45 +02:00
435847891c ggml: cache sin/cos for RoPE (llama/4908) 2024-01-14 00:11:45 +02:00
182f290808 metal : remove old API (llama/4919)
ggml-ci
2024-01-14 00:11:45 +02:00
447dfc11fc metal : disable log for loaded kernels (llama/4794) 2024-01-14 00:11:45 +02:00
9aa9f3b84e gguf : fix potential infinite for-loop (llama/4600)
Co-authored-by: Bernhard Gstrein <gstrein@informatik.uni-freiburg.de>
2024-01-14 00:11:44 +02:00
396ebd1e80 metal : refactor kernel loading code (llama/4794)
* metal : detect more GPU families

* metal : refactor kernel loading

* metal : set kernel family requirements

* metal : fix kernel init + fix compile options

* metal : take into account simdgroup reduction support

* metal : print only skipped kernels

* metal : fix check for simdgroup reduction support

* metal : check for Metal 3

* metal : free allocations

* metal : normalize encoder:setComputePipelineStatus calls

ggml-ci

* metal : fix Metal3 family check

ggml-ci

* metal : check for simdgroup matrix mul. feature

ggml-ci
2024-01-14 00:11:44 +02:00
12490f4398 CUDA: faster q8_0 -> f16 dequantization (llama/4895) 2024-01-14 00:11:44 +02:00
db078a9ba8 talk-llama : add optional CLI arg to set the bot name (#1764) 2024-01-13 20:51:35 +02:00
a13a7da5ad examples : add python example for transcription (#1744)
* rebase and add simple python interface

* moved python files to examples/python
2024-01-13 19:37:18 +02:00
519f8e8684 whisper : load the model into multiple buffers of max size 1GB (#1763) 2024-01-13 17:47:40 +02:00
40ae0962f4 talk-llama : sync llama.cpp 2024-01-12 22:04:51 +02:00
1560288048 sync : ggml 2024-01-12 21:56:50 +02:00
1ad6fafd91 backend_sched : fix assignments
ggml-ci
2024-01-12 21:55:42 +02:00
70840aed5f llama : ggml-backend integration (llama/4766)
* llama : ggml-backend integration

* ggml-backend : add names to buffers

* fix unmap after loading

* batched-bench : add tensor_split param

* llama : check for null tensor_split

* ggml-backend : increase GGML_MAX_BACKENDS

* improve graph splitting, partial fix for --no-kv-offload

* cuda : add ggml-backend split buffer support

* cuda : do not create buffer types for devices that don't exist (fixes usage without CUDA devices available)

* ggml : fix null backend dereference (llama/4807)

* ggml : fix null backend dereference

* ggml : also check ggml_backend_is_cpu

* test-backend-ops : check buffer allocation failures

* llama : add cparam (split_mode) and command line argument (--split-mode, -sm) to configure the split mode (none, layer or row)

* ggml : fix mul_mat_id work size

* llama : rewrite session kv load/set without graphs

* minor

* llama : only initialize used backends, free backends on context free

* llama : abort ctx if cuda backend init fails

* llama : rewrite lora with ggml-backend and compute on CPU

ggml-ci

* llama : only map to a backend buffer the region of the file mapping containing the tensors used in the buffer

* opencl : add ggml-backend buffer type

* cuda : only use batched_cublas with batched mat muls (fixes fp16 tg perf)

* llama : on Metal, by default offload the full model

ggml-ci

* metal : page align the data ptr (llama/4854)

* Apply suggestions from code review

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

* cuda : fix split buffer free

* address review comments

* llama-bench : add split-mode parameter

* fix whitespace

* opencl : fix double initialization

* server : add --split-mode parameter

* use async copy and compute to improve multi-gpu performance

ggml-ci

* use async memcpys to copy the graph outputs to the CPU

* fix opencl

* use a host buffer for the cpu compute buffer for faster copies to the gpu

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
2024-01-12 21:55:42 +02:00
b24d18feb9 CUDA: fix softmax compile for old CUDA versions (llama/4862) 2024-01-12 21:55:41 +02:00
3fa98f4395 Importance Matrix calculation (llama/4861)
* imatrix: 1st version

* imatrix: WIP

* Cleanup

* Update examples/imatrix/imatrix.cpp

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-01-12 21:55:41 +02:00
d05b7ee90e models : make all scripts to be POSIX Compliant (#1725)
* download-coreml-model: make it POSIX-compliant

* download-ggml-model: posix compliant (2nd)

* minor edit

* forgot to add newline

* generate-coreml-interface: far more straightforward

* generate-coreml-model: done with the posix thingy

* typo

* Update download-ggml-model.sh

* fix

* fix typo

* another fix

* Update download-coreml-model.sh

* Update download-ggml-model.sh

* Update download-coreml-model.sh
2024-01-12 14:11:04 +02:00
6dcee35129 ggml : fix 32-bit ARM compat for IQ2_XS (#1758)
* ggml : fix 32-bit ARM compat

* ggml : fix fix

* ggml : fix fix fix
2024-01-12 14:02:30 +02:00
5cb345f5e9 go : add SetInitialPrompt method to bindings (#1753) 2024-01-12 13:44:50 +02:00
fbcb52d3cd server : add more parameters to server api (#1754)
* feat(server): add more parameters to server api

* fix(server): reset params to original parsed values for each request
2024-01-12 13:42:52 +02:00
6b01e3fedd whisper : fix segment length with params.no_timestamps == true 2024-01-12 13:37:38 +02:00
f7908f9bb8 params : don't compute timestamps when not printing them (#1755) 2024-01-12 13:24:38 +02:00
00b7a4be02 talk-llama : sync llama.cpp 2024-01-11 22:10:10 +02:00
04b0a768b8 swift : remove local ggml.h reference 2024-01-11 22:00:12 +02:00
87670425f2 swift : track ggml release branch 2024-01-11 21:57:40 +02:00
32e71a1861 sync : ggml 2024-01-11 21:54:17 +02:00
9c857cf280 sync : llama.cpp 2024-01-11 21:50:01 +02:00
97b12212dd ggml : SOTA 2-bit quants (add IQ2_XS) (llama/4856)
* iq2_xs: basics

* iq2_xs: this should have been in the basics

* iq2_xs: CUDA and scalar CPU works

* iq2_xs: WIP Metal

* iq2_xs: Metal now works

* iq2_xs: working, but dog slow, ARM_NEON dot product

* iq2_xs: better ARM_NEON dot product

We are now at 19.5 t/s for TG-128 and 61 t/s for PP-512 when
running on the CPU.

* iq2_xs: AVX2 dot product - 19.5 t/s

* iq2_xs: faster AVX2 dit product

21.4 t/s for TG-128, 59.2 t/s for PP-512.
The latter is 2x compared to the previous version.

* iq2_xs: had forgotten to delete iq2-data.h

* Add llama enum for IQ2_XS

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 21:50:01 +02:00
9fa34d79ec metal : put encoder debug group behind a define (llama/4873) 2024-01-11 21:50:01 +02:00
a0a64a19dd metal : improve dequantize precision to match CPU (llama/4836)
ggml-ci
2024-01-11 21:50:01 +02:00
bbc23611fa ggml : fix vld1q_s8_x4 32-bit compat (llama/4828)
* ggml : fix vld1q_s8_x4 32-bit compat

ggml-ci

* ggml : fix 32-bit ARM compat (cont)

ggml-ci
2024-01-11 21:50:01 +02:00
e9783a1fb4 CUDA: faster softmax via shared memory + fp16 math (llama/4742) 2024-01-11 21:50:01 +02:00
9e0cc28792 metal : fix deprecation warning (ggml/690) 2024-01-11 21:50:00 +02:00
73072a7c73 ggml : remove ggml_cpy_inplace and ggml_cont_inplace (ggml/693) 2024-01-11 21:50:00 +02:00
a8ba1262ff metal : wrap each operation in debug group (ggml/690) 2024-01-11 21:50:00 +02:00
e66a9a7806 ggml : change GGML_MAX_NAME at compile time (ggml/682)
* change GGML_MAX_NAME to 128

* allow controlling the value of GGML_MAX_NAME through external macro definitions
2024-01-11 21:50:00 +02:00
338442d773 Fix execlp call (ggml/689)
NULL can be an integer constant expression with the value zero, in this case the behavior would be undefined because of an incorrect type being passed to the variable arguments.
2024-01-11 21:50:00 +02:00
10651bddf6 SOTA 2-bit quants (llama/4773)
* iq2_xxs: basics

* iq2_xxs: scalar and AVX2 dot products

Needed to change Q8_K to have quants in the -127...127 range,
else the IQ2_XXS AVX implementation becomes very awkward.
The alternative would have been to use Q8_0 instead. Perhaps
I'll change later, for now this is what we have.

* iq2_xxs: ARM_NEON dot product

Somehow strangely slow (112 ms/token).

* iq2_xxs: WIP Metal

Dequantize works, something is still wrong with the
dot product.

* iq2_xxs: Metal dot product now works

We have
PP-512 = 475 t/s
TG-128 = 47.3 t/s

Not the greatest performance, but not complete garbage either.

* iq2_xxs: slighty faster dot product

TG-128 is now 48.4 t/s

* iq2_xxs: slighty faster dot product

TG-128 is now 50.9 t/s

* iq2_xxs: even faster Metal dot product

TG-128 is now 54.1 t/s.

Strangely enough, putting the signs lookup table
into shared memory has a bigger impact than the
grid values being in shared memory.

* iq2_xxs: dequantize CUDA kernel - fix conflict with master

* iq2_xxs: quantized CUDA dot product (MMVQ)

We get TG-128 = 153.1 t/s

* iq2_xxs: slightly faster CUDA dot product

TG-128 is now at 155.1 t/s.

* iq2_xxs: add to llama ftype enum

* iq2_xxs: fix MoE on Metal

* Fix missing MMQ ops when on hipBLAS

I had put the ggml_supports_mmq call at the wrong place.

* Fix bug in qequantize_row_iq2_xxs

The 0.25f factor was missing.
Great detective work by @ggerganov!

* Fixing tests

* PR suggestion

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
2024-01-11 21:50:00 +02:00