sync : ggml#3193

Merged

ggerganov merged 33 commits into

masterfrom

sync-ggml-25-05-27

May 27, 2025

Commits on May 27, 2025

sync : ggml
ggerganov
committed
Vulkan: Add f32 accumulator support to quantized mul mat to fix GLM4 32B incoherence (llama/13607)

0cc4m
authored and
ggerganov
committed
sycl : Overcoming workaround for mmap() allocation on Windows (llama/13482)

s-Nick
authored and
ggerganov
committed
metal : fix typo in FA kernel comments (llama/13651)
ggerganov
committed
sycl: disable reorder for sycl mulmat (llama/13536)

sgeor255
authored and
ggerganov
committed
CUDA: skip fully masked-out KV in FA vec kernel (llama/13584)

JohannesGaessler
authored and
ggerganov
committed
vulkan: fix warnings (llama/13626)

netrunnereve
authored and
ggerganov
committed
musa: Upgrade MUSA SDK version to rc4.0.1 and use mudnn::Unary::IDENTITY op to accelerate D2D memory copy (llama/13647)

authored and
ggerganov
committed
ggml : add ggml_gelu_erf() (llama/13667)

ngxson
authored and
ggerganov
committed
opencl: fix couple crashes (llama/12795)

Henry Linjamäki
authored and
ggerganov
committed
opencl: Add support for multiple devices (llama/12622)

Henry Linjamäki
authored and
ggerganov
committed
SYCL: Avoid using with SYCL-Graph for unsupported nodes (llama/13587)

Ewan Crawford
authored and
ggerganov
committed
sycl : Remove waits from function calls (llama/13702)

s-Nick
authored and
ggerganov
committed
use LOG_WARN to replace `std::cerr` (llama/13657)

foldl
authored and
ggerganov
committed
vulkan: Disable coopmat/coopmat2/bfloat extensions if glslc doesn't support it (llama/13696)

jeffbolznv
authored and
ggerganov
committed
vulkan: support CPY from any type to itself (llama/13695)

jeffbolznv
authored and
ggerganov
committed
ggml : fix the order of ggml_unary_op (llama/13718)

ngxson
authored and
ggerganov
committed
CANN: Support MUL_MAT_ID for q8_0 and q4_0 (llama/13705)

noemotiovon
authored and
ggerganov
committed
CUDA: fix race condition in FA vector kernels (llama/13742)

JohannesGaessler
authored and
ggerganov
committed
ggml : add ggml_gelu_erf() CUDA kernel (llama/13719)

ngxson
authored and
ggerganov
committed
ggml-cpu : set openmp wait time if not set (llama/13758)

slaren
authored and
ggerganov
committed
SYCL: revert "sycl: simplify bin_bcast_kernel (ggml/13383)" (llama/13752)

qnixsynapse
authored and
ggerganov
committed
CANN: Add the basic supports of Flash Attention kernel (llama/13627)

shibizhao
authored and
ggerganov
committed
vulkan: mark IM2COL as supporting non-contig (llama/13783)

jeffbolznv
authored and
ggerganov
committed
sycl: Add more debug prints (llama/13640)

Rbiessy
authored and
ggerganov
committed
SYCL: Add non contiguous support in RMS_NORM and NORM kernels (llama/13611)

qnixsynapse
authored and
ggerganov
committed
cuda : avoid cuGetErrorString (llama/13791)
ggerganov
committed
ggml : allow CUDA graphs when using pipeline parallelism (llama/13814)

slaren
authored and
ggerganov
committed
ggml-cpu: x86 feature detection is specific to x86 (llama/13811)

ckastner
authored and
ggerganov
committed
ggml : riscv: add xtheadvector support (llama/13720)

xctan
authored and
ggerganov
committed
sync : ggml
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed
sync : fix builds - musa, ruby
ggerganov
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

sync : ggml#3193

sync : ggml#3193
ggerganov merged 33 commits into
masterfrom
sync-ggml-25-05-27

Commits on May 27, 2025

sync : ggml

Vulkan: Add f32 accumulator support to quantized mul mat to fix GLM4 32B incoherence (llama/13607)

sycl : Overcoming workaround for mmap() allocation on Windows (llama/13482)

metal : fix typo in FA kernel comments (llama/13651)

sycl: disable reorder for sycl mulmat (llama/13536)

CUDA: skip fully masked-out KV in FA vec kernel (llama/13584)

vulkan: fix warnings (llama/13626)

musa: Upgrade MUSA SDK version to rc4.0.1 and use mudnn::Unary::IDENTITY op to accelerate D2D memory copy (llama/13647)

ggml : add ggml_gelu_erf() (llama/13667)

opencl: fix couple crashes (llama/12795)

opencl: Add support for multiple devices (llama/12622)

SYCL: Avoid using with SYCL-Graph for unsupported nodes (llama/13587)

sycl : Remove waits from function calls (llama/13702)

use LOG_WARN to replace `std::cerr` (llama/13657)

vulkan: Disable coopmat/coopmat2/bfloat extensions if glslc doesn't support it (llama/13696)

vulkan: support CPY from any type to itself (llama/13695)

ggml : fix the order of ggml_unary_op (llama/13718)

CANN: Support MUL_MAT_ID for q8_0 and q4_0 (llama/13705)

CUDA: fix race condition in FA vector kernels (llama/13742)

ggml : add ggml_gelu_erf() CUDA kernel (llama/13719)

ggml-cpu : set openmp wait time if not set (llama/13758)

SYCL: revert "sycl: simplify bin_bcast_kernel (ggml/13383)" (llama/13752)

CANN: Add the basic supports of Flash Attention kernel (llama/13627)

vulkan: mark IM2COL as supporting non-contig (llama/13783)

sycl: Add more debug prints (llama/13640)

SYCL: Add non contiguous support in RMS_NORM and NORM kernels (llama/13611)

cuda : avoid cuGetErrorString (llama/13791)

ggml : allow CUDA graphs when using pipeline parallelism (llama/13814)

ggml-cpu: x86 feature detection is specific to x86 (llama/13811)

ggml : riscv: add xtheadvector support (llama/13720)

sync : ggml

talk-llama : sync llama.cpp

sync : fix builds - musa, ruby