Pull Requests ggml-org/llama.cpp

opencl: fix two issues on flash attention for Adreno a7x ggml OpenCL

#25697 opened 2026-07-15 04:47 by wanghqc

common : recognize 'thinking' as reasoning alias in chat diff analyzer testing

#25695 opened 2026-07-15 03:08 by michaelneale

sycl: fix row calculation when K_QUANTS_PER_ITERATION is 1 ggml SYCL

#25690 opened 2026-07-14 22:38 by malsbat

ggml-backend-meta: do not scale view offsets internal to a split unit ggml

#25689 opened 2026-07-14 21:40 by Frozenlock

ggml-backend-meta: fix device and buffer type names ggml

#25688 opened 2026-07-14 21:38 by Frozenlock

ggml: force single thread on wasi ggml

#25686 opened 2026-07-14 19:50 by MendyBerger

vulkan: Support quantized concat Vulkan ggml

#25684 opened 2026-07-14 19:34 by jeffbolznv

Add PyTorch dequantization API to gguf-py conversion

#25681 opened 2026-07-14 19:20 by woct0rdho

speculative-simple : support MTP speculative decoding examples

#25679 opened 2026-07-14 19:01 by pdhinaka

cuda : relax tensor contiguity requirements for quantized concat testing ggml CUDA

#25678 opened 2026-07-14 18:51 by fairydreaming

llama-context : sync pending async copies before clearing embd_seq

#25676 opened 2026-07-14 17:35 by o7si

server: add disaggregated prompt prefill workers documentation testing server ggml

#25675 opened 2026-07-14 16:53 by felix314159

tokenize : drop --stdin mutual-exclusion check examples

#25672 opened 2026-07-14 16:35 by angt

rpc : reject invalid top-level graph nodes in graph_compute examples ggml

#25670 opened 2026-07-14 16:31 by liminfei-amd

CUDA: ssm-scan use element strides instead of bytes ggml CUDA

#25669 opened 2026-07-14 16:23 by leonardHONG

vulkan: do not enable MMVQ for speculative-decode steps on AMD Vulkan ggml

#25666 opened 2026-07-14 14:25 by SnoopsDev

et-backend: Q8_0 matrix-engine GEMM kernel for prefill ggml

#25661 opened 2026-07-14 13:52 by RehanQasim-dev

et-backend: F32 decode GEMV rewrite for register-resident vecdot ggml

#25660 opened 2026-07-14 13:52 by RehanQasim-dev

et-backend: F16 decode GEMV rewrite for register-resident vecdot ggml

#25659 opened 2026-07-14 13:52 by RehanQasim-dev

et-backend: Q4_0 matrix-engine GEMM rewrite for prefill ggml

#25658 opened 2026-07-14 13:52 by RehanQasim-dev

[SYCL] fix UT big error of get_rows Q2_K, Q4_K, Q5_K ggml merge ready SYCL

#25656 opened 2026-07-14 13:10 by arthw

[SYCL] support conv2d_dw kernel type is fp16 ggml merge ready SYCL

#25653 opened 2026-07-14 10:18 by arthw

metal : fix null-pipeline crash for F16 src1 mul_mat/mul_mat_id ggml Apple Metal

#25648 opened 2026-07-14 05:33 by QuintinShaw

metal : add F16 support for ADD/SUB/MUL/DIV ggml Apple Metal

#25645 opened 2026-07-14 01:53 by QuintinShaw

common : skip empty implicit default preset

#25643 opened 2026-07-14 01:21 by hogeheer499-commits

ui: fix thinking menu never appearing in single-model mode server/ui

#25637 opened 2026-07-13 19:44 by ServeurpersoCom

llama : fix null pointer dereference by checking for rotation buffer in K-cache shifts

#25636 opened 2026-07-13 19:44 by engeldlgado

CUDA: XOR swizzle flash attn K,V smem fp16 tiles testing ggml CUDA

#25635 opened 2026-07-13 19:23 by ynankani

cuda: extract Q1_0 elements via __byte_perm ggml CUDA

#25628 opened 2026-07-13 16:01 by dfriehs

requierments: Update in requirements/requirements-compare-llama-bench.txt

#25627 opened 2026-07-13 15:37 by Hemansh2633B