Code

sync : llama.cpp#1113

Code

Merged

ggerganov merged 28 commits intomasterggml-org/ggml:masterfrom sync-llama.cpp-25-02-12ggml-org/ggml:sync-llama.cpp-25-02-12Copy head branch name to clipboardFeb 12, 2025

+973-699

Commits

Commits on Feb 12, 2025

• metal : use residency set for other platforms (llama/11648)

  

  jhen0409

  authored and

  ggerganov

  committed
  0752eaf

  Copy full SHA for 0752eaf

• HIP: force max threads per block to be 1024 (llama/11621)

  Show description for 4f37b52
  

  fxzjshm

  authored and

  ggerganov

  committed
  4f37b52

  Copy full SHA for 4f37b52

• CUDA: non-contiguous (RMS) norm support (llama/11659)

  Show description for 387edc9
  

  JohannesGaessler

  and

  ggerganov

  committed
  387edc9

  Copy full SHA for 387edc9

• CUDA: support for mat. mul. with ne03 != ne13 (llama/11656)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  a50d222

  Copy full SHA for a50d222

• metal : adjust support conditions for norm operators (llama/11671)

  Show description for e841c6f
  ggerganov

  committed
  e841c6f

  Copy full SHA for e841c6f

• metal : avoid breaking build when metal API predates TARGET_OS_VISION (llama/11690)

  Show description for 85d428d
  

  charles-dyfis-net

  authored and

  ggerganov

  committed
  85d428d

  Copy full SHA for 85d428d

• vulkan: use smaller combined allocations to avoid fragmentation (llama/11551)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  6cd96d9

  Copy full SHA for 6cd96d9

• vulkan: initial support for IQ4_XS quantization (llama/11501)

  

  remyoudompheng

  authored and

  ggerganov

  committed
  8444353

  Copy full SHA for 8444353

• vulkan: optimize coopmat2 iq2/iq3 callbacks (llama/11521)

  Show description for b23fc86
  

  jeffbolznv

  authored and

  ggerganov

  committed
  b23fc86

  Copy full SHA for b23fc86

• ggml : fix LoongArch compile error with 128-bit SIMD (llama/11701)

  

  junchao-loongson

  authored and

  ggerganov

  committed
  12a4046

  Copy full SHA for 12a4046

• SYCL: Adjust support condition for norm operators (llama/11674)

  Show description for bd19b23
  

  qnixsynapse

  authored and

  ggerganov

  committed
  bd19b23

  Copy full SHA for bd19b23

• ggml : optimize and build warning fix for LoongArch (llama/11709)

  Show description for 2bd5ccb
  

  MQ-mengqing

  authored and

  ggerganov

  committed
  2bd5ccb

  Copy full SHA for 2bd5ccb

• SYCL: remove XMX info from print devices (llama/11712)

  

  qnixsynapse

  authored and

  ggerganov

  committed
  98b3824

  Copy full SHA for 98b3824

• vulkan: print shared memory size (llama/11719)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  e9961a7

  Copy full SHA for e9961a7

• CUDA: fix min. version for movmatrix (llama/11751)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  fdf1349

  Copy full SHA for fdf1349

• ggml: Fix data race in ggml threadpool (llama/11736)

  Show description for 75fc2c3
  

  kkontny

  authored and

  ggerganov

  committed
  75fc2c3

  Copy full SHA for 75fc2c3

• vulkan: account for lookup tables when checking shared memory size (llama/11502)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  1b3c1d2

  Copy full SHA for 1b3c1d2

• vulkan: add environment variable GGML_VK_PREFER_HOST_MEMORY to avoid VRAM allocation (llama/11592)

  

  wbruna

  authored and

  ggerganov

  committed
  e9afd20

  Copy full SHA for e9afd20

• vulkan: Make Vulkan optional at runtime (#11493). (llama/11494)

  Show description for 453aaee
  

  2 peopleauthored and

  ggerganov

  committed
  453aaee

  Copy full SHA for 453aaee

• fix: typos in documentation files (llama/11791)

  Show description for 04b22f0
  

  maximevtush

  authored and

  ggerganov

  committed
  04b22f0

  Copy full SHA for 04b22f0

• CUDA: use arch list for compatibility check (llama/11775)

  Show description for 7f8cf75
  

  2 peopleauthored and

  ggerganov

  committed
  7f8cf75

  Copy full SHA for 7f8cf75

• Fix #11802: Compile bug - RegQueryValueExA changed to RegQueryValueEx (llama/11803)

  Show description for 4b00e83
  

  sheldonrobinson

  authored and

  ggerganov

  committed
  4b00e83

  Copy full SHA for 4b00e83

• CUDA: fix CUDART_VERSION checks (llama/11821)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  501b77b

  Copy full SHA for 501b77b

• ggml-cpu: Fix duplicate MATMUL_INT8 (llama/11817)

  Show description for 0b897da
  

  ownia

  authored and

  ggerganov

  committed
  0b897da

  Copy full SHA for 0b897da

• ggml : fix multi-threaded clamp_f32 (llama/11824)

  Show description for b669f7a
  

  Burton2000

  authored and

  ggerganov

  committed
  b669f7a

  Copy full SHA for b669f7a

• cleanup: fix compile warnings associated with gnu_printf (llama/11811)

  

  bandoti

  authored and

  ggerganov

  committed
  d893024

  Copy full SHA for d893024

• HIP: Switch to std::vector in rocblas version check (llama/11820)

  

  IMbackK

  authored and

  ggerganov

  committed
  cb120dc

  Copy full SHA for cb120dc

• sync : llama.cpp

  Show description for 93ceeb8
  ggerganov

  committed
  93ceeb8

  Copy full SHA for 93ceeb8