sync : ggml and llama.cpp#2429

Merged

ggerganov merged 64 commits intomasterfrom

Sep 24, 2024

Commits on Sep 24, 2024

tests: add gradient tests for all backends (ggml/932)

JohannesGaessler
authored and
ggerganov
committed
vulkan: correctly report support for OP_CONT (ggml/946)

smeso
authored and
ggerganov
committed
vulkan: add dryrun support to sin and cos ops (ggml/947)

smeso
authored and
ggerganov
committed
vulkan: fix compilation with GGML_VULKAN_DEBUG=ON (ggml/948)

smeso
authored and
ggerganov
committed
Threadpool: take 2 (llama/8672)

authored and
ggerganov
committed
llama : support RWKV v6 models (llama/8980)

committed
ggml : add pthread includes on FreeBSD (llama/9258)

yuri@FreeBSD
authored and
ggerganov
committed
Fix DMMV dequantization (llama/9279)

OuadiElfarouki
authored and
ggerganov
committed
ggml : AVX2 support for Q4_0_8_8 (llama/8713)

Srihari-mcw
authored and
ggerganov
committed
cuda : fix defrag with quantized KV (llama/9319)

slaren
authored and
ggerganov
committed
ggml-quants : ternary packing for TriLMs and BitNet b1.58 (llama/8151)

compilade
authored and
ggerganov
committed
Improve Vulkan shader build system (llama/9239)

mtavenrath
authored and
ggerganov
committed
ggml : fix missing `cpu_set_t` on emscripten (llama/9336)

ngxson
authored and
ggerganov
committed
ggml : always check bounds on get_rows operations (llama/9354)

slaren
authored and
ggerganov
committed
metal : update support condition for im2col + fix warning (llama/0)
ggerganov
committed
examples : add null threadpool args where needed (ggml/0)
ggerganov
committed
ggml/examples: add backend support for numerical optimization (ggml/949)

committed
add check malloc result on device (llama/9346)

authored and
ggerganov
committed
cuda : fix FA Q src index (1 -> 0) (llama/9374)
ggerganov
committed
Overlap cmdbuffer creation and cmdbuffer execution in Vulkan backend by submitting smaller cmdbuffers early. (llama/9118)

mtavenrath
authored and
ggerganov
committed
CUDA: fix variable name conflict for Windows build (llama/9382)

JohannesGaessler
authored and
ggerganov
committed
ggml : vector length agnostic SVE support (llama/9290)

Vithulep
and
ggerganov
committed
rpc : fix segfault with nkvo (llama/9389)

authored and
ggerganov
committed
metal : fix compile warning with GGML_METAL_NDEBUG (llama/0)
ggerganov
committed
sycl : update support conditions (llama/9394)

Alcpz
authored and
ggerganov
committed
musa: remove Clang builtins mapping (llama/9421)

yeahdongcn
authored and
ggerganov
committed
CUDA: fix --split-mode row race condition (llama/9413)

JohannesGaessler
authored and
ggerganov
committed
cann: Fix error when running a non-exist op (llama/9424)

bachelor-dou
authored and
ggerganov
committed
riscv : modify Makefile and add a RISCV_VECT to print log info (llama/9442)

Tameem-10xE
authored and
ggerganov
committed
cann: Add host buffer type for Ascend NPU (llama/9406)

bachelor-dou
authored and
ggerganov
committed
cmake : use list(APPEND ...) instead of set() + dedup linker (llama/9463)

ggerganov
and
Xarbirus
committed
ggml : ggml_type_name return "NONE" for invalid values (llama/9458)

ykhrustalev
authored and
ggerganov
committed
cmake : try to fix sycl+intel build (llama/9487)

Xarbirus
authored and
ggerganov
committed
cmake : correct order of sycl flags (llama/9497)

Xarbirus
authored and
ggerganov
committed
common : reimplement logging (llama/9418)
ggerganov
committed
metal : handle zero-sized allocs (llama/9466)
ggerganov
committed
ggml : IQ4_NL sgemm + Q4_0 AVX optimization (llama/9422)

netrunnereve
authored and
ggerganov
committed
cmake : do not hide GGML options + rename option (llama/9465)
ggerganov
committed
ggml : link MATH_LIBRARY not by its full path (llama/9339)

Xarbirus
authored and
ggerganov
committed
threadpool : skip polling for unused threads (llama/9461)

max-krasnyansky
authored and
ggerganov
committed
ggml : fix n_threads_cur initialization with one thread (llama/9538)

authored and
ggerganov
committed
CUDA: fix sum.cu compilation for CUDA < 11.7 (llama/9562)

JohannesGaessler
authored and
ggerganov
committed
ggml : fix trailing whitespace (llama/0)
ggerganov
committed
ggml : fix builds (llama/0)
ggerganov
committed
ggml : refactoring (llama/#0)
ggerganov
committed
examples : adapt to ggml.h changes (ggml/0)
ggerganov
committed
Update CUDA graph on scale change plus clear nodes/params (llama/9550)

agray3
authored and
ggerganov
committed
ggml-alloc : fix list of allocated tensors with GGML_ALLOCATOR_DEBUG (llama/9573)

slaren
authored and
ggerganov
committed
RWKV v6: RWKV_WKV op CUDA implementation (llama/9454)

MollySophia
authored and
ggerganov
committed
CUDA: enable Gemma FA for HIP/Pascal (llama/9581)

JohannesGaessler
authored and
ggerganov
committed
Fix merge error in #9454 (llama/9589)

MollySophia
authored and
ggerganov
committed
musa: enable building fat binaries, enable unified memory, and disable Flash Attention on QY1 (MTT S80) (llama/9526)

yeahdongcn
authored and
ggerganov
committed
Revert "[SYCL] fallback mmvq (ggml/9088)" (llama/9579)

Akarshan Biswas
authored and
ggerganov
committed
metal : use F32 prec for K*Q in vec FA (llama/9595)
ggerganov
committed
ggml : AVX512 gemm for Q4_0_8_8 (llama/9532)

Srihari-mcw
and
ggerganov
committed
threads: improve ggml_barrier scaling with large number of threads (llama/9598)

max-krasnyansky
authored and
ggerganov
committed
cuda: add q8_0->f32 cpy operation (llama/9571)

Nekotekina
authored and
ggerganov
committed
threads: fix msvc build without openmp (llama/9615)

max-krasnyansky
authored and
ggerganov
committed
log : add CONT level for continuing previous log entry (llama/9610)
ggerganov
committed
ggml : add AVX512DQ requirement for AVX512 builds (llama/9622)

EZForever
authored and
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed
sync : ggml
ggerganov
committed
ggml : add ggml-cpu-impl.h (skip) (#0)
ggerganov
committed
make : remove "talk" target until updated
ggerganov
committed