Pull Requests ggerganov/llama.cpp

Add cohere2moe to llama-vocab for TINY_AYA

#24601 opened 2026-06-14 04:37 by bartowski1182

[SYCL] support OPs: conv_2d, conv_2d_dw, conv2d_transpose documentation examples ggml SYCL

#24600 opened 2026-06-14 04:30 by arthw

server: add system message prefix feature examples python server

#24599 opened 2026-06-14 02:15 by coder543

ci: fix vulkan docker images Vulkan ggml

#24595 opened 2026-06-13 21:15 by Kononnable

ci : use CUDA label for cuda backend devops

#24594 opened 2026-06-13 20:58 by CISC

spec: support eagle3 for qwen3.5 & 3.6 model examples server

#24593 opened 2026-06-13 20:48 by ruixiang63

hexagon: support for op-trace (fine-grain tracing of HVX/HMX/DMA events) script python ggml Hexagon

#24592 opened 2026-06-13 20:47 by max-krasnyansky

llama : suppress misleading Gemma4Assistant error during memory fitting

#24590 opened 2026-06-13 19:26 by leotm

HIP: use hipBLAS for dense prefill on gfx900, keep MMQ for MoE Nvidia GPU ggml

#24588 opened 2026-06-13 17:39 by DEV-DUFORD

vulkan: add iq4_nl support back to FA Vulkan ggml

#24585 opened 2026-06-13 16:50 by jeffbolznv

[SYCL] add to support pool_1d, move pool_1d/2d code to pool.cpp/hpp documentation ggml merge ready SYCL

#24584 opened 2026-06-13 16:23 by arthw

vulkan: support all backend tests for SQR/SQRT/SIN/COS/CLAMP/LEAKY_RELU/NORM testing Nvidia GPU Vulkan ggml WebGPU

#24582 opened 2026-06-13 15:22 by jeffbolznv

vulkan: Support gated_delta_net with S_v=16 Vulkan ggml

#24581 opened 2026-06-13 15:09 by jeffbolznv

vulkan: support more CONCAT types testing Vulkan ggml

#24579 opened 2026-06-13 15:02 by jeffbolznv

[SYCL]fix reorder function crash:GGML_ASSERT(block_num_y % num_subgroups ==0) examples ggml merge ready SYCL

#24578 opened 2026-06-13 14:50 by arthw

ui: (demo) access server remotely via webrtc examples server/ui

#24577 opened 2026-06-13 14:45 by ngxson

ggml: optimize concat op by replacing per-element memcpy with row-level memcpy ggml

#24575 opened 2026-06-13 14:07 by sirohikartik

CI: Replace flake8-no-print with flake8-debug and pin repos to hashes

#24572 opened 2026-06-13 12:41 by jpodivin

CUDA: Add conv3d. Nvidia GPU ggml

#24569 opened 2026-06-13 11:24 by Sero1000

EXPERIMENT: meta: key external view cache by backend context ggml

#24566 opened 2026-06-13 08:45 by nycdubliner

[fattn-tune] Add Blackwell MMA config Nvidia GPU ggml

#24565 opened 2026-06-13 07:04 by yaohengxu

[SYCL] Enhance set_rows to support q1_0, mxfp4, nvfp4 documentation ggml merge ready SYCL

#24564 opened 2026-06-13 06:48 by arthw

CUDA: don't route RDNA3.5 flash attention to the rocWMMA kernel Nvidia GPU ggml

#24562 opened 2026-06-13 03:39 by liminfei-amd

CUDA/HIP: chunked MFMA prefill kernel for GATED_DELTA_NET (CDNA) testing Nvidia GPU ggml

#24561 opened 2026-06-13 03:08 by jadenmach2

ggml-alloc : check realloc result in alloc_tensor_range ggml

#24559 opened 2026-06-13 02:03 by ricku777-bear

Fix 24486: TP: allows the usage of 8,9,10 gpus for stepfun

#24554 opened 2026-06-13 01:18 by krampenschiesser

llama: copy tensor_split at model load instead of retaining caller pointer, resolving segfault

#24552 opened 2026-06-13 01:07 by dragonfyre13

llama : disable graph reuse when contexts share memory under SPLIT_MODE_TENSOR

#24549 opened 2026-06-12 23:52 by nycdubliner

Reduce RSS during BF16 GGUF export python

#24548 opened 2026-06-12 23:11 by i386

ggml-cuda: use universal launch bounds for MoE MMVQ kernel Nvidia GPU ggml

#24547 opened 2026-06-12 23:05 by batot1