]>
git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/log
Diego Devesa [Thu, 12 Dec 2024 18:02:49 +0000 (19:02 +0100)]
remove CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS (llama/10797)
other windows build fixes
0cc4m [Thu, 12 Dec 2024 17:36:00 +0000 (18:36 +0100)]
Vulkan: Use improved q4_k and q5_k dequant code in dequant shaders (llama/10798)
0cc4m [Thu, 12 Dec 2024 17:35:37 +0000 (18:35 +0100)]
Vulkan: Add VK_EXT_subgroup_size_control support to ensure full subgroups for coopmats (llama/10721)
* Vulkan: Add VK_EXT_subgroup_size_control support to ensure full subgroups for coopmats
* Fix subgroup size control extension support check
Add accf32 and accf16 checks for coopmats
* Also disable coopmats on amdvlk
Gilad S [Wed, 11 Dec 2024 00:47:21 +0000 (02:47 +0200)]
ggml: load all backends from a user-provided search path (llama/10699)
* feat: load all backends from a user-provided search path
* fix: Windows search path
* refactor: rename `ggml_backend_load_all_in_search_path` to `ggml_backend_load_all_from_path`
* refactor: rename `search_path` to `dir_path`
* fix: change `NULL` to `nullptr`
Co-authored-by: Diego Devesa <redacted>
* fix: change `NULL` to `nullptr`
---------
Co-authored-by: Diego Devesa <redacted>
Jeff Bolz [Tue, 10 Dec 2024 20:23:17 +0000 (14:23 -0600)]
vulkan: request round-to-even for fp16 in im2col/rope_head (llama/10767)
Vulkan doesn't mandate a specific rounding mode, but the shader_float_controls
feature allows rounding mode to be requested if the implementation supports it.
Eve [Tue, 10 Dec 2024 19:33:23 +0000 (19:33 +0000)]
vulkan: dynamic subgroup size for the remaining k quants (llama/10745)
* q5_k
q4_k
q3_k
q2_k
q6_k multi row example
* revert as multi row isnt faster for k quants
Andreas Kieslinger [Tue, 10 Dec 2024 17:23:24 +0000 (18:23 +0100)]
CUDA: rename macros to avoid conflicts with WinAPI (llama/10736)
* Renames NVIDIA GPU-architecture flags to avoid name clashes with WinAPI. (e.g. CC_PASCAL, GPU architecture or WinAPI pascal compiler flag?)
* Reverts erroneous rename in SYCL-code.
* Renames GGML_CUDA_MIN_CC_DP4A to GGML_CUDA_CC_DP4A.
* Renames the rest of the compute capability macros for consistency.
Jeff Bolz [Tue, 10 Dec 2024 17:22:20 +0000 (11:22 -0600)]
vulkan: disable spirv-opt for coopmat shaders (llama/10763)
There are some bugs in the 1.3.296 SDK, so disable this. It isn't strictly
necessary anyway.
Add missing dependency on vulkan-shaders-gen, so shaders get recompiled when it
changes.
Fix coopmat support reporting when glslc doesn't support NV_coopmat2.
Daniel Bevenius [Sat, 14 Dec 2024 02:23:08 +0000 (03:23 +0100)]
ggml : remove return from ggml_gallocr_allocate_node (ggml/1048)
This commit removes the return statement from ggml_gallocr_allocate_node
function.
The motivation behind this change is to make the code more readable and
consistent.
Daniel Bevenius [Fri, 13 Dec 2024 07:19:38 +0000 (08:19 +0100)]
ggml : add check for grad_accs (ggml/1046)
* ggml : add check for grad_accs
This commit adds a check for grad_accs in ggml_graph_get_grad and
ggml_graph_get_grad_acc functions. This is necessary to avoid segfaults
when grad_accs is not initialized.
The motivation for this change is that I find it nice to be able to
print out a computation graph using ggml_graph_print but this function
segfaults when grad_accs is not initialized:
```console
(gdb) p g1
$2 = (ggml_cgraph *) 0x7ffff66004b0
(gdb) p *g1
$3 = {size = 2048, n_nodes = 1, n_leafs = 2, nodes = 0x7ffff6600500,
grads = 0x0, grad_accs = 0x0, leafs = 0x7ffff6604500,
visited_hash_set = {size = 4099, used = 0x7ffff6610518,
keys = 0x7ffff6608500}, order = GGML_CGRAPH_EVAL_ORDER_LEFT_TO_RIGHT}
(gdb) p ggml_graph_print(g1)
=== GRAPH ===
n_nodes = 1
Program received signal SIGSEGV, Segmentation fault.
0x0000555555579775 in ggml_graph_get_grad
(cgraph=0x7ffff66004b0,node=0x7ffff6600340)
at /ggml/ggml/src/ggml.c:5990
5990 return igrad != GGML_HASHSET_FULL &&
ggml_bitset_get(cgraph->visited_hash_set.used, igrad) ?
cgraph->grads[igrad] : NULL;
```
* squash! ggml : add check for grad_accs
Fix the check in ggml_graph_get_grad. The check was incorrectly using
cgraph->grad_accs instead of cgraph->grads.
Georgi Gerganov [Tue, 10 Dec 2024 15:19:09 +0000 (17:19 +0200)]
common : remove old types
ggml-ci
Johannes Gäßler [Mon, 9 Dec 2024 19:07:12 +0000 (20:07 +0100)]
CUDA: fix shared memory access condition for mmv (llama/10740)
Jeff Bolz [Mon, 9 Dec 2024 07:24:01 +0000 (01:24 -0600)]
vulkan: fix compile warnings (llama/10731)
stduhpf [Sun, 8 Dec 2024 18:19:19 +0000 (19:19 +0100)]
Vulkan: fix NaN in tanh.comp with AMD proprietary driver on Windows (llama/10723)
* Vulkan: fix NaN in tanh.comp
* Faster NaN-free tanh
Jeff Bolz [Sun, 8 Dec 2024 08:05:55 +0000 (02:05 -0600)]
vulkan: compile a test shader in cmake to check for coopmat2 support (llama/10713)
Georgi Gerganov [Sat, 7 Dec 2024 16:38:15 +0000 (18:38 +0200)]
ggml : disable iq4_nl interleave size 8 (llama/10709)
ggml-ci
Djip007 [Sat, 7 Dec 2024 12:37:50 +0000 (13:37 +0100)]
ggml : refactor online repacking (llama/10446)
* rename ggml-cpu-aarch64.c to .cpp
* reformat extra cpu backend.
- clean Q4_0_N_M and IQ4_0_N_M
- remove from "file" tensor type
- allow only with dynamic repack
- extract cpu extra bufts and convert to C++
- hbm
- "aarch64"
- more generic use of extra buffer
- generalise extra_supports_op
- new API for "cpu-accel":
- amx
- aarch64
* clang-format
* Clean Q4_0_N_M ref
Enable restrict on C++
* add op GGML_OP_MUL_MAT_ID for Q4_0_N_M with runtime repack
* added/corrected control on tensor size for Q4 repacking.
* Update ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp
Co-authored-by: Georgi Gerganov <redacted>
* Update ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp
Co-authored-by: Georgi Gerganov <redacted>
* add debug logs on repacks.
---------
Co-authored-by: Georgi Gerganov <redacted>
0cc4m [Sat, 7 Dec 2024 09:24:15 +0000 (10:24 +0100)]
Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing (llama/10597)
* Vulkan: Implement VK_KHR_cooperative_matrix support in the matrix matrix multiplication shader
* Improve performance with better q4_k and q5_k dequant and store unrolling
* Add Vulkan MUL_MAT and MUL_MAT_ID accumulator precision selection
* Rework mulmat shader selection and compilation logic, avoid compiling shaders that won't get used by device
* Vulkan: Implement accumulator switch for specific mul mat mat shaders
* Vulkan: Unroll more loops for more mul mat mat performance
* Vulkan: Add VK_AMD_shader_core_properties2 support to read Compute Unit count for split_k logic
* Disable coopmat support on AMD proprietary driver
* Remove redundant checks
* Add environment variable GGML_VK_DISABLE_COOPMAT to disable VK_KHR_cooperative_matrix support
* Fix rebase typo
* Fix coopmat2 MUL_MAT_ID pipeline selection
Robert Ormandi [Sat, 7 Dec 2024 07:55:01 +0000 (01:55 -0600)]
metal : Extend how Llama.cpp locates metal resources (llama/10676)
* metal : Extend how Llama.cpp locates metal resources (llama/10675)
* It searches the resource file in the directory where the current
binary is located as well.
* Resolves symbolic links.
Rationale:
When we plug this dependency into a Bazel build and run it in the
context of Bazel (e.g. testing):
* the execution directory is often very different from where the files
are located and no direct control over this (Bazel sandboxing),
* the Bazel sandbox often use symbolic links to make files available.
With this patch, we can have the resource file added to the target,
can build and run tests in the context of Bazel.
* Update ggml/src/ggml-metal/ggml-metal.m
Co-authored-by: Georgi Gerganov <redacted>
* Update ggml/src/ggml-metal/ggml-metal.m
Co-authored-by: Georgi Gerganov <redacted>
---------
Co-authored-by: Georgi Gerganov <redacted>
Jeff Bolz [Thu, 5 Dec 2024 19:15:05 +0000 (13:15 -0600)]
vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)
KITAITI Makoto [Wed, 18 Dec 2024 09:00:50 +0000 (18:00 +0900)]
ruby : Add no_speech_thold (#2641)
* Remove Whisper::Model.[]
* Fix Whisper::Model::URI#request
* Make Whisper::Context#initialize accept pre-converted model name
* Use downloading pre-converted model feature for testing
* Update README
* Remove unnecessary task
* Move whisper/model.rb -> whisper/model/uri.rb
* Update document comment of Whisper::Context#initialize
* Don't show download progress when not tty
* Pass String to raise
* Use cache model file if download fails
* Add test for auto download
* Specify required Ruby version
* Fix a typo
* Remove unnecessary flags
* Initialize Whisper::Params#diarize explicitely
* Remove redundant code from README for simplicity
* Add Whisper::Params#no_speech_thold attribute
* Add test for Whisper::Params#no_speech_thold
crummyh [Wed, 18 Dec 2024 06:43:48 +0000 (00:43 -0600)]
stream : improve consistency in README (#2642)
Karthick [Tue, 17 Dec 2024 17:15:47 +0000 (22:45 +0530)]
whisper : support no_speech_thold (#2625)
* Implement no_speech_thold
no_speech_thold functionality is on par with OpenAI's whisper
* Addressed review comments
Karthick [Tue, 17 Dec 2024 17:07:08 +0000 (22:37 +0530)]
whisper : add single-timestamp logic (#2629)
* Fix hallucinations during silence
When the predicted tokens end with a single timestamp the the entire 30 segment should be considered as done, to avoid hallucinations for the remaining part of segment.
This behaviour is on par with openai's whisper. Refer to logic related to `single_timestamp_ending` in https://github.com/openai/whisper/blob/main/whisper/transcribe.py
* Accept review comments related to formatting.
Co-authored-by: Georgi Gerganov <redacted>
---------
Co-authored-by: Georgi Gerganov <redacted>
crummyh [Tue, 17 Dec 2024 17:05:35 +0000 (11:05 -0600)]
readme : fix typo (#2637)
Georgi Gerganov [Tue, 17 Dec 2024 16:34:32 +0000 (18:34 +0200)]
cmake : fix "amd64" processor string (#2638)
gn64 [Mon, 16 Dec 2024 10:34:38 +0000 (19:34 +0900)]
vulkan : fix soft_max.comp division by zero (#2633)
This change prevents a division by zero error when p.KY is 0.
Georgi Gerganov [Mon, 16 Dec 2024 06:57:04 +0000 (08:57 +0200)]
common : add cstdio header
Georgi Gerganov [Sun, 15 Dec 2024 19:55:36 +0000 (21:55 +0200)]
stream : update build instructions
Thamster [Sat, 14 Dec 2024 15:25:53 +0000 (10:25 -0500)]
android : fix build and ci (#2624)
* Adding missing CMakeLists.txt include for ggm-cpu needed by whisper.android
* attempt to re-enable CI for JNI android
---------
Co-authored-by: Your Name <redacted>
Michael Rienstra [Thu, 12 Dec 2024 16:02:00 +0000 (08:02 -0800)]
models : fix typo in download-ggml-model.sh (#2623)
Introduced in #2589
KITAITI Makoto [Mon, 9 Dec 2024 11:17:50 +0000 (20:17 +0900)]
ruby : Sync whisper.cpp and model download feature (#2617)
* Use C++17
* Add test for Pathname of model
* Make Whisper::Context#initialize accept Pathname
* Add shorthand for pre-converted models
* Update documents
* Add headings to API section in README [skip ci]
* Remove unused function
* Don't care about no longer included file
* Cosmetic fix
* Use conditional get when get model files
Georgi Gerganov [Mon, 9 Dec 2024 09:30:16 +0000 (11:30 +0200)]
scripts : update to new build system
Georgi Gerganov [Sun, 8 Dec 2024 21:09:26 +0000 (23:09 +0200)]
devops : add cmake
Georgi Gerganov [Sun, 8 Dec 2024 21:07:29 +0000 (23:07 +0200)]
devops : update make commands
Georgi Gerganov [Sun, 8 Dec 2024 21:04:26 +0000 (23:04 +0200)]
ggml : remove old files (skip) (#0)
Georgi Gerganov [Sun, 8 Dec 2024 20:48:25 +0000 (22:48 +0200)]
ggml : sync remnants (skip) (#0)
Georgi Gerganov [Sun, 8 Dec 2024 20:48:14 +0000 (22:48 +0200)]
scripts : remove amx from sync
Georgi Gerganov [Sun, 8 Dec 2024 13:52:57 +0000 (15:52 +0200)]
ci : disable freeBSD builds [no ci]
Georgi Gerganov [Sun, 8 Dec 2024 13:48:14 +0000 (15:48 +0200)]
readme : update build instructions
Georgi Gerganov [Sun, 8 Dec 2024 13:36:01 +0000 (15:36 +0200)]
ci : disable CUDA and Android builds
Georgi Gerganov [Sun, 8 Dec 2024 11:35:35 +0000 (13:35 +0200)]
ci : disable Obj-C build + fixes
Georgi Gerganov [Fri, 6 Dec 2024 13:34:53 +0000 (15:34 +0200)]
make : shim cmake
Georgi Gerganov [Thu, 5 Dec 2024 12:30:33 +0000 (14:30 +0200)]
talk-llama : sync llama.cpp
Georgi Gerganov [Thu, 5 Dec 2024 12:29:18 +0000 (14:29 +0200)]
sync : ggml
Diego Devesa [Wed, 4 Dec 2024 13:45:40 +0000 (14:45 +0100)]
ggml : add predefined list of CPU backend variants to build (llama/10626)
* ggml : add predefined list of CPU backend variants to build
* update CPU dockerfiles
Diego Devesa [Wed, 4 Dec 2024 13:40:44 +0000 (14:40 +0100)]
ggml-cpu : fix HWCAP2_I8MM value (llama/10646)
Jeff Bolz [Wed, 4 Dec 2024 07:28:59 +0000 (01:28 -0600)]
vulkan: Implement "fast divide" (mul+shift) for unary ops like copy (llama/10642)
Nicolò Scipione [Wed, 4 Dec 2024 01:29:20 +0000 (02:29 +0100)]
SYCL : Move to compile time oneMKL interface backend selection for NVIDIA backend (llama/10584)
* [SYCL] Move to Compile Time backend selection on oneMKL Interface for NVIDIA backend
Move to compile time selection to backend to avoid latency at run time.
Add it to all mkl gemm calls and only for NVIDIA backend.
Signed-off-by: nscipione <redacted>
* Formatting
* Address PR comments to increase readibility
---------
Signed-off-by: nscipione <redacted>
Frankie Robertson [Wed, 4 Dec 2024 00:41:37 +0000 (02:41 +0200)]
Avoid using __fp16 on ARM with old nvcc (llama/10616)
Jeff Bolz [Tue, 3 Dec 2024 19:29:54 +0000 (13:29 -0600)]
vulkan: optimize and reenable split_k (llama/10637)
Use vector loads when possible in mul_mat_split_k_reduce. Use split_k
when there aren't enough workgroups to fill the shaders.
PAB [Wed, 4 Dec 2024 08:19:30 +0000 (09:19 +0100)]
ggml: add `GGML_SET` Metal kernel + i32 CPU kernel (ggml/1037)
* implemented cpu kernel
* add i32 test cases in test-backend-ops
* typedef `ggml_metal_kargs_set`
* implemented `kernel_set`
* memcpy
PAB [Tue, 3 Dec 2024 19:20:04 +0000 (20:20 +0100)]
ggml : add `GGML_PAD_REFLECT_1D` operation (ggml/1034)
* ggml_pad_reflect_1d defined in header
* implemented on CPU
* called the forward pass
* impl Metal kernel
* added Metal kernel
* added OP_PAD_REFLECT_1D in test-backend-ops.cpp
* add test-pad-reflect-1d test case
* test case support multiple backend
Georgi Gerganov [Tue, 3 Dec 2024 18:29:32 +0000 (20:29 +0200)]
files : remove make artifacts
Georgi Gerganov [Tue, 3 Dec 2024 18:25:37 +0000 (20:25 +0200)]
common : fix compile warning
ggml-ci
Diego Devesa [Tue, 3 Dec 2024 18:22:12 +0000 (20:22 +0200)]
ggml : move AMX to the CPU backend (llama/10570)
ggml : automatic selection of best CPU backend (llama/10606)
Georgi Gerganov [Tue, 3 Dec 2024 09:52:33 +0000 (11:52 +0200)]
metal : small-batch mat-mul kernels (llama/10581)
* metal : small-batch mat-mul kernels
ggml-ci
* metal : add rest of types
ggml-ci
* metal : final adjustments
ggml-ci
* metal : add comments
ggml-ci
Akarshan Biswas [Mon, 2 Dec 2024 07:04:11 +0000 (12:34 +0530)]
SYCL: Fix and switch to GGML_LOG system instead of fprintf (llama/10579)
* Switched to GGML_LOG
* Fix missing semicolon
Adrien Gallouët [Sat, 30 Nov 2024 17:13:18 +0000 (18:13 +0100)]
ggml-cpu: replace AArch64 NEON assembly with intrinsics in ggml_gemv_q4_0_4x4_q8_0() (llama/10567)
Signed-off-by: Adrien Gallouët <redacted>
Eve [Sat, 30 Nov 2024 07:00:02 +0000 (07:00 +0000)]
vulkan: Dynamic subgroup size support for Q6_K mat_vec (llama/10536)
* subgroup 64 version with subgroup add. 15% faster
scalable version
tested for subgroup sizes 16-128
* check for subgroup multiple of 16 and greater than 16
* subgroup sizes are always a power of 2 (https://github.com/KhronosGroup/GLSL/issues/45)
* force 16 sequential threads per block
* make 16 subgroup size a constant
Georgi Gerganov [Fri, 29 Nov 2024 14:25:39 +0000 (16:25 +0200)]
ggml : fix I8MM Q4_1 scaling factor conversion (llama/10562)
ggml-ci
Shupei Fan [Fri, 29 Nov 2024 13:49:02 +0000 (21:49 +0800)]
ggml-cpu: fix typo in gemv/gemm iq4_nl_4_4 (llama/10580)
Alberto Cabrera Pérez [Fri, 29 Nov 2024 12:38:45 +0000 (12:38 +0000)]
sycl : offload of get_rows set to 0 (llama/10432)
Alberto Cabrera Pérez [Fri, 29 Nov 2024 09:49:43 +0000 (09:49 +0000)]
sycl : Reroute permuted mul_mats through oneMKL (llama/10408)
This PR fixes the failing MUL_MAT tests for the sycl backend.
Chenguang Li [Fri, 29 Nov 2024 06:46:55 +0000 (14:46 +0800)]
CANN: RoPE operator optimization (llama/10563)
* [cann] RoPE operator optimization
* [CANN]Code Formatting
---------
Co-authored-by: noemotiovon <redacted>
Jeff Bolz [Fri, 29 Nov 2024 06:18:02 +0000 (00:18 -0600)]
vulkan: get the first command buffer submitted sooner (llama/10499)
This is an incremental improvement over #9118 to get work to the GPU a bit
sooner. The first part is to start with a smaller number of nodes before
the first submit, and ramp it up to the current 100 nodes/submit. The
second part is to reduce the dryrun overhead for all the nodes that just
need to request descriptor space.
With these changes I get around 1-2% speedup on RTX 4070 combined with my
old Haswell-era CPU.
Georgi Gerganov [Thu, 28 Nov 2024 18:46:40 +0000 (20:46 +0200)]
ggml : remove redundant copyright notice + update authors
Georgi Gerganov [Thu, 28 Nov 2024 12:56:37 +0000 (14:56 +0200)]
ggml : fix row condition for i8mm kernels (llama/10561)
ggml-ci
Georgi Gerganov [Thu, 28 Nov 2024 12:56:23 +0000 (14:56 +0200)]
cmake : fix ARM feature detection (llama/10543)
ggml-ci
Shupei Fan [Thu, 28 Nov 2024 12:52:03 +0000 (20:52 +0800)]
ggml-cpu: support IQ4_NL_4_4 by runtime repack (llama/10541)
* ggml-cpu: support IQ4_NL_4_4 by runtime repack
* ggml-cpu: add __ARM_FEATURE_DOTPROD guard
Sergio López [Thu, 28 Nov 2024 11:51:38 +0000 (12:51 +0100)]
kompute : improve backend to pass test_backend_ops (llama/10542)
* kompute: op_unary: reject unsupported parameters
Signed-off-by: Sergio Lopez <redacted>
* kompute: softmax: implement ALiBi support
Signed-off-by: Sergio Lopez <redacted>
* kompute: rope: implement neox and phi3 support
Signed-off-by: Sergio Lopez <redacted>
* kompute: op_mul_mat_q4_k permutted support
Signed-off-by: Sergio Lopez <redacted>
* kompute: op_mul_mat_[q4_0|q4_1|q8_0] permutted support
Signed-off-by: Sergio Lopez <redacted>
* kompute: op_mul_mat_f16 permutted support
Signed-off-by: Sergio Lopez <redacted>
* kompute: op_mul_mat_q6_k permutted support
Signed-off-by: Sergio Lopez <redacted>
---------
Signed-off-by: Sergio Lopez <redacted>
leo-pony [Thu, 28 Nov 2024 07:25:24 +0000 (15:25 +0800)]
CANN: Fix SOC_TYPE compile bug (llama/10519)
* CANN: Fix the bug build fail on Ascend310P under two cases:
1) Manual specify SOC_TYPE
2) Under some unusual compile environment
* Update the cann backend News content: Support F16 and F32 data type model for Ascend 310P NPU.
* fix CANN compile fail bug: the assert in ascend kernel function doesn't supportted on some CANN version
Chenguang Li [Thu, 28 Nov 2024 06:24:46 +0000 (14:24 +0800)]
CANN: ROPE operator optimization (llama/10540)
* [cann] ROPE operator optimization
Co-authored-by: noemotiovon <redacted>
uvos [Wed, 27 Nov 2024 16:10:08 +0000 (17:10 +0100)]
Add some minimal optimizations for CDNA (llama/10498)
* Add some minimal optimizations for CDNA
* ggml_cuda: set launch bounds also for GCN as it helps there too
Georgi Gerganov [Wed, 27 Nov 2024 09:22:14 +0000 (11:22 +0200)]
metal : fix group_norm support condition (llama/0)
Jeff Bolz [Wed, 27 Nov 2024 07:32:54 +0000 (01:32 -0600)]
vulkan: define all quant data structures in types.comp (llama/10440)
Jeff Bolz [Wed, 27 Nov 2024 07:30:27 +0000 (01:30 -0600)]
vulkan: Handle GPUs with less shared memory (llama/10468)
There have been reports of failure to compile on systems with <= 32KB
of shared memory (e.g. #10037). This change makes the large tile size
fall back to a smaller size if necessary, and makes mul_mat_id fall
back to CPU if there's only 16KB of shared memory.
Jeff Bolz [Wed, 27 Nov 2024 07:21:59 +0000 (01:21 -0600)]
vulkan: further optimize q5_k mul_mat_vec (llama/10479)
Jeff Bolz [Wed, 27 Nov 2024 07:08:54 +0000 (01:08 -0600)]
vulkan: skip integer div/mod in get_offsets for batch_idx==0 (llama/10506)
Jeff Bolz [Wed, 27 Nov 2024 07:00:50 +0000 (01:00 -0600)]
vulkan: optimize Q2_K and Q3_K mul_mat_vec (llama/10459)
R0CKSTAR [Tue, 26 Nov 2024 16:00:41 +0000 (00:00 +0800)]
mtgpu: Add MUSA_DOCKER_ARCH in Dockerfiles && update cmake and make (llama/10516)
Signed-off-by: Xiaodong Ye <redacted>
Jeff Bolz [Tue, 26 Nov 2024 15:45:05 +0000 (09:45 -0600)]
vulkan: fix group_norm (llama/10496)
Fix bad calculation of the end of the range. Add a backend test that
covers the bad case (taken from stable diffusion).
Fixes https://github.com/leejet/stable-diffusion.cpp/issues/439.
Georgi Gerganov [Tue, 26 Nov 2024 12:18:08 +0000 (14:18 +0200)]
cmake : enable warnings in llama (llama/10474)
* cmake : enable warnings in llama
ggml-ci
* cmake : add llama_get_flags and respect LLAMA_FATAL_WARNINGS
* cmake : get_flags -> ggml_get_flags
* speculative-simple : fix warnings
* cmake : reuse ggml_get_flags
ggml-ci
* speculative-simple : fix compile warning
ggml-ci
Charles Xu [Tue, 26 Nov 2024 11:37:05 +0000 (12:37 +0100)]
ggml-cpu: cmake add arm64 cpu feature check for macos (llama/10487)
* ggml-cpu: cmake add arm64 cpu feature check for macos
* use vmmlaq_s32 for compile option i8mm check
Shanshan Shen [Tue, 26 Nov 2024 10:08:37 +0000 (18:08 +0800)]
CANN: Improve the Inferencing Performance for Ascend NPU Device (llama/10454)
* improve inferencing performance for ascend npu.
Co-authored-by: Frank Mai <redacted>
* some modification after review
* some modifications after review
* restore some modifications
* restore some modifications
---------
Co-authored-by: shanshan shen <redacted>
Co-authored-by: Frank Mai <redacted>
Chenguang Li [Tue, 26 Nov 2024 09:31:05 +0000 (17:31 +0800)]
CANN: RoPE and CANCAT operator optimization (llama/10488)
Co-authored-by: noemotiovon <redacted>
Junil Kim [Tue, 26 Nov 2024 01:47:20 +0000 (10:47 +0900)]
vulkan: Fix a vulkan-shaders-gen arugment parsing error (llama/10484)
The vulkan-shaders-gen was not parsing the --no-clean argument correctly.
Because the previous code was parsing the arguments which have a value only
and the --no-clean argument does not have a value, it was not being parsed
correctly. This commit can now correctly parse arguments that don't have values.
Georgi Gerganov [Mon, 25 Nov 2024 19:49:31 +0000 (21:49 +0200)]
metal : enable mat-vec kernels for bs <= 4 (llama/10491)
Diego Devesa [Mon, 25 Nov 2024 18:30:06 +0000 (19:30 +0100)]
llama : accept a list of devices to use to offload a model (llama/10497)
* llama : accept a list of devices to use to offload a model
* accept `--dev none` to completely disable offloading
* fix dev list with dl backends
* rename env parameter to LLAMA_ARG_DEVICE for consistency
Diego Devesa [Mon, 25 Nov 2024 14:13:39 +0000 (15:13 +0100)]
ggml : add support for dynamic loading of backends (llama/10469)
* ggml : add support for dynamic loading of backends
---------
Co-authored-by: Georgi Gerganov <redacted>
Georgi Gerganov [Mon, 25 Nov 2024 13:08:04 +0000 (15:08 +0200)]
metal : minor code formatting
Diego Devesa [Sat, 23 Nov 2024 13:41:12 +0000 (14:41 +0100)]
ggml : do not use ARM features not included in the build (llama/10457)
leo-pony [Fri, 22 Nov 2024 06:07:20 +0000 (14:07 +0800)]
CANN: Support Ascend310P to accelerate F32 and F16 Model (llama/10216)
* CANN Support Ascend310P to accelerate F32 and F16 Model
* Add compile option soc type macro ASCEND_310P to ggml-cann lib
* Remove unused code
* Remove the ascend soc_type hard code compile option in CMakelist.txt
Diego Devesa [Thu, 21 Nov 2024 17:18:50 +0000 (18:18 +0100)]
cuda : optimize argmax (llama/10441)
* cuda : optimize argmax
* remove unused parameter
ggml-ci
* fixup : use full warps
ggml-ci
* Apply suggestions from code review
Co-authored-by: Johannes Gäßler <redacted>
* fix ub
* ggml : check ne00 <= INT32_MAX in argmax and argsort
---------
Co-authored-by: Johannes Gäßler <redacted>
Jeff Bolz [Wed, 20 Nov 2024 19:47:36 +0000 (13:47 -0600)]
vulkan: predicate max operation in soft_max shaders/soft_max (llama/10437)
Fixes #10434
Jeff Bolz [Wed, 20 Nov 2024 07:40:18 +0000 (01:40 -0600)]
vulkan: copy iq4_nl LUT into shared memory (llama/10409)
Jeff Bolz [Wed, 20 Nov 2024 07:11:00 +0000 (01:11 -0600)]
vulkan: further optimize mul_mat_vec using larger loads (llama/10387)
* vulkan: Use pipeline_robustness to disable robustness in mul_mat_vec.
Add some early returns for nonexistent rows in mul_mat_vec shaders. These
can only be hit when dispatching a 2D grid of workgroups. Fix the logic
for the 2D grid of workgroups to round up.
Enable the pipeline robustness extension if it's available, and use it to
disable robustness for these pipelines. The instructions to do the bounds
checking contend for the same ALU resources as the bit twiddling dequant
instructions.
* vulkan: Add GLSL structure aliases for quant types to allow larger loads
In Vulkan it's not possible to cast pointer types, so instead you have to
declare an aliased binding for the memory with a different type. This
commit adds aliases for the quant formats using 16b ints, and in a few
places where the struct size is a multiple of 4 also using 32b ints.
Currently only q4_k's aliases are used, but others will be used in
subsequent commits.
* vulkan: use larger loads in q5_k and q6_k shaders.
Similar to the optimization I did in q4_k recently, this vectorizes some loads
and reduces the number of bit twiddling instructions.
* vulkan: use larger K step per iteration in mul_mat_vec.
Add vec4 dequantization functions, and use them to do K=8 per iteration in
mul_mat_vec. This uses 16b loads for the quant values and 128b loads for B
which helps reduce the load on the memory system.
The K_PER_ITER==2 logic is still there, just for F16/F32, and really only
because they support unaligned sizes.
Tweak the num_iters/unrolling logic to be simpler and catch a couple missed
unrolling opportunities.
haopeng [Tue, 19 Nov 2024 20:10:31 +0000 (04:10 +0800)]
add cmake rvv support (llama/10411)
mahorozte [Tue, 3 Dec 2024 13:11:43 +0000 (21:11 +0800)]
CUDA: remove unnecessary warp reduce in FA (ggml/1032)
* kqmax_new_j in every thread within warp is same after operate at line 199,this reduce can be omit
* same problem in vec32
---------
Co-authored-by: ZhaoXiaoYu <redacted>
PAB [Mon, 2 Dec 2024 18:27:24 +0000 (19:27 +0100)]
feat: add `GGML_UNARY_OP_ARGMAX` Metal kernel (ggml/1019)
* implemented argmax kernel
* tpig -> tgpig
* change to strides
* contiguous assertions
* kernel working and tested
* argmax simd parallel implementation
* added 2 new tests for argmax in test-backend-ops
* cosmit
* added 3 tests cases for perf eval
* add test_argmax in make_test_cases_perf
* Update test-backend-ops.cpp
Co-authored-by: Diego Devesa <redacted>
---------
Co-authored-by: Diego Devesa <redacted>