sync : ggml#2639

Merged

ggerganov merged 36 commits intomasterfrom

sync-ggml-24-12-17-0

Dec 18, 2024

Commits on Dec 17, 2024

vulkan: Add VK_NV_cooperative_matrix2 support for mul_mat and flash attention (llama/10206)

jeffbolznv
authored and
ggerganov
committed
metal : Extend how Llama.cpp locates metal resources (llama/10676)

ormandi
and
ggerganov
committed
Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing (llama/10597)

0cc4m
authored and
ggerganov
committed
ggml : refactor online repacking (llama/10446)

Djip007
and
ggerganov
committed
ggml : disable iq4_nl interleave size 8 (llama/10709)
ggerganov
committed
vulkan: compile a test shader in cmake to check for coopmat2 support (llama/10713)

jeffbolznv
authored and
ggerganov
committed
Vulkan: fix NaN in tanh.comp with AMD proprietary driver on Windows (llama/10723)

stduhpf
authored and
ggerganov
committed
vulkan: fix compile warnings (llama/10731)

jeffbolznv
authored and
ggerganov
committed
CUDA: fix shared memory access condition for mmv (llama/10740)

JohannesGaessler
authored and
ggerganov
committed
common : remove old types
ggerganov
committed
ggml : add check for grad_accs (ggml/1046)

danbev
authored and
ggerganov
committed
ggml : remove return from ggml_gallocr_allocate_node (ggml/1048)

danbev
authored and
ggerganov
committed
vulkan: disable spirv-opt for coopmat shaders (llama/10763)

jeffbolznv
authored and
ggerganov
committed
CUDA: rename macros to avoid conflicts with WinAPI (llama/10736)

aendk
authored and
ggerganov
committed
vulkan: dynamic subgroup size for the remaining k quants (llama/10745)

netrunnereve
authored and
ggerganov
committed
vulkan: request round-to-even for fp16 in im2col/rope_head (llama/10767)

jeffbolznv
authored and
ggerganov
committed
ggml: load all backends from a user-provided search path (llama/10699)

authored and
ggerganov
committed
Vulkan: Add VK_EXT_subgroup_size_control support to ensure full subgroups for coopmats (llama/10721)

0cc4m
authored and
ggerganov
committed
Vulkan: Use improved q4_k and q5_k dequant code in dequant shaders (llama/10798)

0cc4m
authored and
ggerganov
committed
remove CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS (llama/10797)

slaren
authored and
ggerganov
committed
CUDA: faster non-contiguous concat (llama/10760)

authored and
ggerganov
committed
ggml : Fix compilation issues on ARM platform when building without fp16 (llama/10811)

kkontny
authored and
ggerganov
committed
SYCL: Reduce most of the compiler warnings (llama/10748)

authored and
ggerganov
committed
vulkan: small mul_mat_vec optimizations (llama/10665)

netrunnereve
authored and
ggerganov
committed
Fix crash caused by ggml_backend_load_all when launching on Android Activity (llama/10812)

authored and
ggerganov
committed
Introducing experimental OpenCL backend with support for Qualcomm Adreno GPUs (llama/10693)

authored and
ggerganov
committed
llama : add Qwen2VL support + multimodal RoPE (llama/10361)

HimariO
and
ggerganov
committed
rwkv6: add wkv6 support for Vulkan backend (llama/10829)

authored and
ggerganov
committed
vulkan: bugfixes for small subgroup size systems + llvmpipe test (llama/10809)

netrunnereve
authored and
ggerganov
committed
ggml : update ggml_backend_cpu_device_supports_op (llama/10867)
ggerganov
committed
sync : ggml
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed

Commits on Dec 18, 2024