Pull Requests ggerganov/llama.cpp

llama : do not register GPU devices when n_gpu_layers == 0

#31090 opened 2026-10-24 11:29 by Rocke1001feller

llama : respect no_alloc in all memory caches

#31087 opened 2026-10-24 10:10 by ggerganov

model: DSV4.1 flash model testing ggml mtmd CUDA conversion

#31086 opened 2026-10-24 10:09 by am17an

ggml : possible fix for the radix TOP_K thread timing bug testing Vulkan ggml CUDA

#31084 opened 2026-10-24 09:11 by sanmai

ggml-cpu : gate RVV intrinsics on __riscv_vector (#22159) ggml

#31082 opened 2026-10-24 08:59 by Xeonacid

ggml : link Threads::Threads as PUBLIC on ggml-base (#31007) ggml

#31080 opened 2026-10-24 07:36 by Fogyvishnu

vulkan : fix out-of-bounds NVFP4 UE4M3 scale lookup Vulkan ggml

#31079 opened 2026-10-24 07:27 by flyingdoog

sycl : Try PkgConfig for LevelZero includes / libs ggml SYCL

#31076 opened 2026-10-24 04:26 by mctylr-gh

metal : faster MXFP4 mat-vec, full threadgroups for unary/binary/GLU ggml Apple Metal

#31075 opened 2026-10-24 04:25 by am17an

kv-cache : keep rotation constants on the GPU testing

#31073 opened 2026-10-24 03:38 by anton-iermolenko

hexagon: generic dims collapse/reshape for unary, binary and matmul ops merge ready

#31071 opened 2026-10-24 03:28 by max-krasnyansky

unicode : add continuation check lambda

#31070 opened 2026-10-24 02:57 by danbev

nix: expose a SYCL package nix devops

#31066 opened 2026-10-24 01:47 by Titaniumtown

sycl : bound reorder temp memory to 32 MiB ggml SYCL

#31062 opened 2026-10-23 21:37 by KyleHagy

ggml-cpu: unroll x86 quantized vec_dot accumulator chains (Q4_0, Q4_K, Q5_K, IQ3_S, IQ3_XXS) ggml

#31061 opened 2026-10-23 20:13 by silicium369-spec

grammar: fix GBNF {m,n} repetition uint64 underflow (max<min) causing…

#31060 opened 2026-10-23 20:08 by thesecguy45

jinja : fix select and selectattr argument handling testing jinja parser

#31059 opened 2026-10-23 19:37 by JoeyPatricio

improve gemm4xN Q8_0 on AVX2 ggml

#31058 opened 2026-10-23 19:30 by aconz2

hexagon: KV-head team split for multicore HMX flash attention testing ggml Hexagon

#31055 opened 2026-10-23 17:48 by ebateni

Hexagon: IM2COL improvement - block based HVX parallelization testing ggml Hexagon

#31050 opened 2026-10-23 16:23 by rjtokenring

cmake: Fix pkg-config module missing requirements when using system GGML build mtmd

#31048 opened 2026-10-23 15:43 by amyspark

CUDA: linearize MoE data layout for MMQ/MMF ggml CUDA

#31047 opened 2026-10-23 15:41 by JohannesGaessler

vulkan : enable MMQ with mixed signedness integer dots Vulkan ggml

#31045 opened 2026-10-23 14:11 by birkskyum

tools : reject truncated GGUF tensor data in merge examples merge ready

#31044 opened 2026-10-23 13:54 by shreya-11733

ggml-cpu/amx: use a constant-initialized tile config avoid dead store elimination ggml

#31042 opened 2026-10-23 12:36 by medns

common/autoparser : exact bitmask permutation for mixed required/optional tagged tool args testing

#31040 opened 2026-10-23 11:58 by tugayoktayokay

MTP draft with a trimmed output head model

#31037 opened 2026-10-23 11:24 by budiony

fix exhausting CPU and MEM caused by dry sequence breaker documentation testing server

#31035 opened 2026-10-23 09:44 by trunk-trick

jinja: fix stack overflow in from_json when chat_template_kwargs cont… testing jinja parser

#31032 opened 2026-10-23 09:11 by Frost-54

Trimmed draft vocab support (1/2) - reduced draft LM head from a user supplied token list documentation testing server

#31031 opened 2026-10-23 08:13 by praneshgo