Code

sync : ggml#2342

Code

Merged

ggerganov merged 70 commits intomasterggml-org/whisper.cpp:masterfrom

syncggml-org/whisper.cpp:syncCopy head branch name to clipboard

Aug 8, 2024

Commits

Commits on Aug 8, 2024

• scripts : sync new files (#0)

  ggerganov

  committed
  05b3d1c

  Copy full SHA for 05b3d1c

• cmake : only enable GGML_NATIVE and x86 flags if not crosscompiling (ggml/885)

  

  iboB

  authored and

  ggerganov

  committed
  4fce5d4

  Copy full SHA for 4fce5d4

• vulkan : initialize vk_buffer_struct members to VK_NULL_HANDLE (ggml/893)

  Show description for f0de462
  

  neobrain

  authored and

  ggerganov

  committed
  f0de462

  Copy full SHA for f0de462

• ggml: add support for float16 input tensors in pooling operations (ggml/895)

  Show description for 52be7d6
  

  2 peopleauthored and

  ggerganov

  committed
  52be7d6

  Copy full SHA for 52be7d6

• ggml : loop tiling optimizations for scalar path (ggml/898)

  Show description for bcf05d5
  

  heshpdx

  authored and

  ggerganov

  committed
  bcf05d5

  Copy full SHA for bcf05d5

• sycl : fix powf call in device code (llama/8368)

  

  Alcpz

  authored and

  ggerganov

  committed
  12e31b1

  Copy full SHA for 12e31b1

• sycl : Reenabled mmvq path for the SYCL Nvidia Backend (llama/8372)

  Show description for 43816f9
  

  Alcpz

  authored and

  ggerganov

  committed
  43816f9

  Copy full SHA for 43816f9

• ggml : add AArch64 optimized GEMV and GEMM Q4 kernels (llama/5780)

  Show description for 0fc4674
  

  Dibakar

  authored and

  ggerganov

  committed
  0fc4674

  Copy full SHA for 0fc4674

• ggml : move sgemm sources to llamafile subfolder (llama/8394)

  Show description for dc9508d
  ggerganov

  committed
  dc9508d

  Copy full SHA for dc9508d

• Use multi_ptr to clean up deprecated warnings (llama/8256)

  

  AidanBeltonS

  authored and

  ggerganov

  committed
  6b733da

  Copy full SHA for 6b733da

• CUDA: optimize and refactor MMQ (llama/8416)

  Show description for aed5f19
  

  JohannesGaessler

  authored and

  ggerganov

  committed
  aed5f19

  Copy full SHA for aed5f19

• cuda : suppress 'noreturn' warn in no_device_code (llama/8414)

  Show description for c7bb780
  

  danbev

  authored and

  ggerganov

  committed
  c7bb780

  Copy full SHA for c7bb780

• ggml : add NVPL BLAS support (ggml/8329) (llama/8425)

  Show description for e298c16
  

  2 peopleauthored and

  ggerganov

  committed
  e298c16

  Copy full SHA for e298c16

• fix the mul_mat_id ut issues (llama/8427)

  Show description for c46ee3e
  

  3 peopleauthored and

  ggerganov

  committed
  c46ee3e

  Copy full SHA for c46ee3e

• ggml : minor naming changes (llama/8433)

  Show description for 33fe98b
  ggerganov

  committed
  33fe98b

  Copy full SHA for 33fe98b

• metal : template-ify some of the kernels (llama/8447)

  Show description for c4b504b
  ggerganov

  committed
  c4b504b

  Copy full SHA for c4b504b

• vulkan : cmake integration (llama/8119)

  Show description for e88d9f6
  

  bandoti

  authored and

  ggerganov

  committed
  e88d9f6

  Copy full SHA for e88d9f6

• Vulkan MMQ Fix (llama/8479)

  Show description for df07dae
  

  0cc4m

  authored and

  ggerganov

  committed
  df07dae

  Copy full SHA for df07dae

• add concat through dim 1/2 (llama/8483)

  Show description for fb2bd6f
  

  airMeng

  authored and

  ggerganov

  committed
  fb2bd6f

  Copy full SHA for fb2bd6f

• Refactor lora adapter support (llama/8332)

  Show description for b946bf8
  

  3 peopleauthored and

  ggerganov

  committed
  b946bf8

  Copy full SHA for b946bf8

• make/cmake: add missing force MMQ/cuBLAS for HIP (llama/8515)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  3c63419

  Copy full SHA for 3c63419

• Add Ascend NPU backend (llama/6035)

  Show description for 83f9da7
  

  2 peopleauthored and

  ggerganov

  committed
  83f9da7

  Copy full SHA for 83f9da7

• cmake : install all ggml public headers (llama/8480)

  Show description for 53b9f09
  

  2 peopleauthored and

  ggerganov

  committed
  53b9f09

  Copy full SHA for 53b9f09

• CUDA: fix partial offloading for ne0 % 256 != 0 (llama/8572)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  c0760e2

  Copy full SHA for c0760e2

• ggml : add friendlier error message to fopen errors (llama/8575)

  Show description for 05a9656
  

  HanClinto

  authored and

  ggerganov

  committed
  05a9656

  Copy full SHA for 05a9656

• ggml : fix quant dot product with odd number of blocks (llama/8549)

  Show description for 027863f
  

  slaren

  and

  ggerganov

  committed
  027863f

  Copy full SHA for 027863f

• gguf : handle null name during init (llama/8587)

  ggerganov

  committed
  a7e1d3a

  Copy full SHA for a7e1d3a

• CUDA: MMQ code deduplication + iquant support (llama/8495)

  Show description for 54d968f
  

  JohannesGaessler

  authored and

  ggerganov

  committed
  54d968f

  Copy full SHA for 54d968f

• ggml: fix compile error for RISC-V (llama/8623)

  

  zqb-all

  authored and

  ggerganov

  committed
  5b28839

  Copy full SHA for 5b28839

• fix scratch size of softmax (llama/8642)

  

  luoyu-intel

  authored and

  ggerganov

  committed
  8899f0b

  Copy full SHA for 8899f0b

• Allow all RDNA2 archs to use sdot4 intrinsic (llama/8629)

  Show description for 6ab790f
  

  jeroen-mostert

  authored and

  ggerganov

  committed
  6ab790f

  Copy full SHA for 6ab790f

• Vulkan IQ4_NL Support (llama/8613)

  Show description for 67a39e8
  

  0cc4m

  authored and

  ggerganov

  committed
  67a39e8

  Copy full SHA for 67a39e8

• sycl : Add support for non-release DPC++ & oneMKL (llama/8644)

  Show description for 5beea92
  

  joeatodd

  authored and

  ggerganov

  committed
  5beea92

  Copy full SHA for 5beea92

• Re-add erroneously removed -fsycl from GGML_EXTRA_LIBS (llama/8667)

  

  joeatodd

  authored and

  ggerganov

  committed
  41c25ae

  Copy full SHA for 41c25ae

• ggml : add and use ggml_cpu_has_llamafile() (llama/8664)

  ggerganov

  committed
  25ae1ae

  Copy full SHA for 25ae1ae

• fix multi-gpu issue on sycl (llama/8554)

  Show description for 95a1378
  

  2 peopleauthored and

  ggerganov

  committed
  95a1378

  Copy full SHA for 95a1378

• ggml: handle ggml_init failure to fix NULL pointer deref (llama/8692)

  Show description for 5ad818e
  

  DavidKorczynski

  authored and

  ggerganov

  committed
  5ad818e

  Copy full SHA for 5ad818e

• ggml : reduce hash table reset cost (llama/8698)

  Show description for 10befde
  

  slaren

  authored and

  ggerganov

  committed
  10befde

  Copy full SHA for 10befde

• ggml : add ggml-aarch64 (ggml/0)

  

  Dibakar

  authored and

  ggerganov

  committed
  b6796f5

  Copy full SHA for b6796f5

• common : handle new quant types (ggml/0)

  ggerganov

  committed
  6cd7612

  Copy full SHA for 6cd7612

• ggml : resolve sync conflicst (ggml/0)

  Show description for 96cf53f
  ggerganov

  committed
  96cf53f

  Copy full SHA for 96cf53f

• ggml : move c parameter comment to ggml_rope_ext (ggml/901)

  Show description for f57c22a
  

  danbev

  authored and

  ggerganov

  committed
  f57c22a

  Copy full SHA for f57c22a

• vulkan : implement Stable Diffusion operators (ggml/904)

  Show description for 887537c
  

  0cc4m

  authored and

  ggerganov

  committed
  887537c

  Copy full SHA for 887537c

• metal : add abort callback (ggml/905)

  

  conradev

  authored and

  ggerganov

  committed
  3f40857

  Copy full SHA for 3f40857

• metal : fix struct name (ggml/912)

  Show description for c626ad5
  ggerganov

  committed
  c626ad5

  Copy full SHA for c626ad5

• ggml : ignore more msvc warnings (ggml/906)

  

  iboB

  authored and

  ggerganov

  committed
  2e4a575

  Copy full SHA for 2e4a575

• feat: Support Moore Threads GPU (llama/8383)

  Show description for ded669b
  

  yeahdongcn

  authored and

  ggerganov

  committed
  ded669b

  Copy full SHA for ded669b

• add conv support (llama/8688)

  

  airMeng

  authored and

  ggerganov

  committed
  9ae2794

  Copy full SHA for 9ae2794

• cuda : organize vendor-specific headers into vendors directory (llama/8746)

  Show description for e6d4565
  

  yeahdongcn

  authored and

  ggerganov

  committed
  e6d4565

  Copy full SHA for e6d4565

• ggml: bugfix: fix the inactive elements is agnostic for risc-v vector (llama/8748)

  Show description for d153915
  

  2 peopleauthored and

  ggerganov

  committed
  d153915

  Copy full SHA for d153915

• Add `TIMESTEP_EMBEDDING` OP (llama/8707)

  Show description for 5eff773
  

  zhentaoyu

  authored and

  ggerganov

  committed
  5eff773

  Copy full SHA for 5eff773

• cann: update cmake (llama/8765)

  

  wangshuai09

  authored and

  ggerganov

  committed
  7f82c97

  Copy full SHA for 7f82c97

• added android implementation of ggml_print_backtrace_symbols (llama/8751)

  Show description for 362030c
  

  2 peopleauthored and

  ggerganov

  committed
  362030c

  Copy full SHA for 362030c

• cuda : fix dmmv cols requirement to 2*GGML_CUDA_DMMV_X (llama/8800)

  Show description for 003db16
  

  slaren

  authored and

  ggerganov

  committed
  003db16

  Copy full SHA for 003db16

• Build: Only include execinfo.h on linux systems that support it (llama/8783)

  Show description for a5550c9
  

  acon96

  authored and

  ggerganov

  committed
  a5550c9

  Copy full SHA for a5550c9

• ggml-cuda: Adding support for unified memory (llama/8035)

  Show description for 997ef81
  

  2 peopleauthored and

  ggerganov

  committed
  997ef81

  Copy full SHA for 997ef81

• Fixing wrong VDR iq4nl value (llama/8812)

  

  OuadiElfarouki

  authored and

  ggerganov

  committed
  94ea73e

  Copy full SHA for 94ea73e

• Fix conversion of unnormalized BF16->BF16 weights (llama/7843)

  Show description for 4996a13
  

  2 peopleauthored and

  ggerganov

  committed
  4996a13

  Copy full SHA for 4996a13

• ggml : reading the runtime sve config of the cpu (llama/8709)

  Show description for d044468
  

  2 peopleauthored and

  ggerganov

  committed
  d044468

  Copy full SHA for d044468

• ggml : fix overflows in elu function (llama/8866)

  Show description for 4898fae
  

  jart

  authored and

  ggerganov

  committed
  4898fae

  Copy full SHA for 4898fae

• ggml : add epsilon as a parameter for group_norm (llama/8818)

  Show description for f26565a
  

  MollySophia

  authored and

  ggerganov

  committed
  f26565a

  Copy full SHA for f26565a

• CUDA: fix padding logic for FP16/FP32 (llama/8884)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  5bc1f65

  Copy full SHA for 5bc1f65

• CUDA/HIP: fix tests/test-backend-ops (llama/8896)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  3b8a579

  Copy full SHA for 3b8a579

• Updated SYCL device filtering (llama/8901)

  Show description for 4d1b494
  

  OuadiElfarouki

  authored and

  ggerganov

  committed
  4d1b494

  Copy full SHA for 4d1b494

• ggml-backend : fix async copy from CPU (llama/8897)

  Show description for 69e9538
  

  slaren

  authored and

  ggerganov

  committed
  69e9538

  Copy full SHA for 69e9538

• sync : ggml

  ggerganov

  committed
  17749ad

  Copy full SHA for 17749ad

• talk-llama : sync llama.cpp

  ggerganov

  committed
  da43d55

  Copy full SHA for da43d55

• build : fix aarch64 (#0)

  ggerganov

  committed
  0504fba

  Copy full SHA for 0504fba

• ci : try to fix FreeBSD (#0)

  ggerganov

  committed
  224c75f

  Copy full SHA for 224c75f

• ci : disable ruby workflow (#0)

  ggerganov

  committed
  bb21cb3

  Copy full SHA for bb21cb3