sync : ggml#2561

Merged

ggerganov merged 55 commits intomasterfrom

Nov 15, 2024

Commits on Nov 15, 2024

scripts : update sync
ggerganov
committed
metal : fix minor string leaks (ggml/1004)

pminev
authored and
ggerganov
committed
cmake : make it possible linking ggml as external lib (ggml/1003)

ykhrustalev
authored and
ggerganov
committed
musa: workaround for Guilty Lockup in cleaning src0 (llama/10042)

yeahdongcn
authored and
ggerganov
committed
ggml: Add POOL2D OP for GPU acceleration to the Vulkan backend in the MobileVLM model. (llama/9763)

cyzero-kim
authored and
ggerganov
committed
llama : refactor model loader with backend registry (llama/10026)

slaren
authored and
ggerganov
committed
ggml : add Q4_0_8_8 RISC-V GEMV and GEMM kernels (llama/10029)

xctan
authored and
ggerganov
committed
ggml : fix memory leaks when loading invalid gguf files (llama/10094)

slaren
authored and
ggerganov
committed
kompute: add backend registry / device interfaces (llama/10045)

slp
authored and
ggerganov
committed
kompute: add mul_mat_q4_k shader (llama/10097)

slp
authored and
ggerganov
committed
ggml : check tensor name lengths in gguf files (llama/10100)

slaren
authored and
ggerganov
committed
llama : fix buffer checks for mamba and rwk (llama/10111)

slaren
authored and
ggerganov
committed
build: fix build error in Windows env with OneAPI setup (llama/10107)

kylo5aby
authored and
ggerganov
committed
ggml : remove ggml_scratch (llama/10121)
ggerganov
committed
vulkan : improve ggml_vk_create_buffer error handling (llama/9898)

FanShupei
authored and
ggerganov
committed
llama : use smart pointers for ggml resources (llama/10117)

slaren
authored and
ggerganov
committed
llama : add simple-chat example (llama/10124)

authored and
ggerganov
committed
metal : minor fixup in FA kernel (llama/10143)
ggerganov
committed
ggml : move CPU backend to a separate file (llama/10144)

slaren
authored and
ggerganov
committed
CANN: adjust backend registry refactor. (llama/10158)

leo-pony
authored and
ggerganov
committed
metal : move dequantize templates to beginning of MSL source (llama/0)
ggerganov
committed
metal : simplify f16 and f32 dequant kernels (llama/0)
ggerganov
committed
cuda : clear error after changing peer access (llama/10153)

slaren
authored and
ggerganov
committed
fix build break on arm64 linux (llama/10166)

snadampal
authored and
ggerganov
committed
ggml : fix q4xx mat mul, increase ggml_aligned_malloc alignment (llama/10167)

slaren
authored and
ggerganov
committed
ggml : fix gelu tables initialization (llama/10172)

slaren
authored and
ggerganov
committed
Q6_K AVX improvements (llama/10118)

netrunnereve
authored and
ggerganov
committed
ggml : fix arch check in bf16_to_fp32 (llama/10164)

slaren
authored and
ggerganov
committed
metal : add quantized FA support (llama/10149)
ggerganov
committed
ggml : adjust is_first_call init value (llama/10193)
ggerganov
committed
metal : fix from ptr buffer name (llama/10189)

slaren
authored and
ggerganov
committed
metal : add BF16 support (llama/8439)
ggerganov
committed
Optimize RWKV6 Operator Naming and Implement Multi-core CPU/ SYCL Acceleration (llama/10133)

committed
fix q4_0_8_8 format for corrupted tokens issue (llama/10198)

authored and
ggerganov
committed
ggml : add ggml-cpu.h to the public headers (llama/10204)

slaren
authored and
ggerganov
committed
metal : optimize FA kernels (llama/10171)
ggerganov
committed
metal : improve clarity (minor) (llama/10171)
ggerganov
committed
metal : opt-in compile flag for BF16 (llama/10218)
ggerganov
committed
ggml : optimize llamafile cpu matrix multiplication for ppc64le (llama/10156)

amritahs-ibm
authored and
ggerganov
committed
ggml: fix zero division in ‘dne’ calculation in CUDA COUNT_EQUAL operator when ‘ne’ is small (#10213)

SongXiaoXi
authored and
ggerganov
committed
metal : hide debug messages from normal log
ggerganov
committed
metal : fix F32 accumulation in FA vec kernel (llama/10232)
ggerganov
committed
metal : fix build and some more comments (llama/10229)
ggerganov
committed
metal : reorder write loop in mul mat kernel + style (llama/10231)
ggerganov
committed
vulkan: Fix newly added tests for permuted mul_mat and 1D im2col (llama/10226)

jeffbolznv
authored and
ggerganov
committed
metal : more precise Q*K in FA vec kernel (llama/10247)
ggerganov
committed
vulkan: Throttle the number of shader compiles during the build step. (llama/10222)

jeffbolznv
authored and
ggerganov
committed
vulkan: Optimize contiguous copies (llama/10254)

jeffbolznv
authored and
ggerganov
committed
sycl : Fixes to broken builds and test-backend-ops (llama/10257)

Alcpz
authored and
ggerganov
committed
sync : ggml
ggerganov
committed
whisper : fix build (#0)
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed
build : fixes
ggerganov
committed
whisper : include ggml-cpu.h (#0)
ggerganov
committed
cmake : fix ppc64 check (#0)
ggerganov
committed