Pull Requests ggerganov/llama.cpp

CI: Fix docker multiarch overwrite devops

#21144 opened 2026-03-29 07:09 by Ts-sound

common: add two-phase graceful reasoning budget termination ...

#21141 opened 2026-03-29 03:03 by zeel2104

grammar: make MAX_REPETITION_THRESHOLD configurable via env var

#21139 opened 2026-03-29 01:12 by vampyrebat

Multi-backend profiler Nvidia GPU Vulkan examples python ggml SYCL Ascend NPU OpenCL Hexagon OpenVINO

#21138 opened 2026-03-29 00:21 by pwilkin

hexagon: dma optimizations (mostly fixing regressions) ggml Hexagon

#21137 opened 2026-03-28 23:48 by max-krasnyansky

CI: Enable CUDA ARM64 runners documentation devops

#21122 opened 2026-03-28 15:26 by ehfd

metal: add opt-in V skip for negligible attention weights ggml Apple Metal

#21119 opened 2026-03-28 13:09 by TheTom

convert: Add compressed-tensors NVFP4 conversion python

#21095 opened 2026-03-28 04:09 by michaelw9999

server/webui: cleanup dual representation approach, simplify to openai-compat examples server

#21090 opened 2026-03-28 00:06 by pwilkin

ggml : add CPU TurboQuant KV cache types (TBQ3_0 / TBQ4_0) testing examples server ggml

#21089 opened 2026-03-27 23:59 by elusznik

[CUDA] Reduce the number of stream-k blocks to reduce the overhead of the flash_attn_stream_k_fixup kernel Nvidia GPU ggml

#21086 opened 2026-03-27 22:14 by gaugarg-nv

common: add bounds check in common_init_result::sampler to prevent segfault on failed model load testing examples

#21082 opened 2026-03-27 18:58 by mtmcp

fix cmake problem to exclude CCAN need more info ggml Ascend NPU

#21075 opened 2026-03-27 16:59 by sunqingn7

ggml-cuda: Add generic NVFP4 MMQ kernel Nvidia GPU python ggml

#21074 opened 2026-03-27 16:50 by michaelw9999

server: (webui) no more gzip compression server/webui examples server

#21073 opened 2026-03-27 16:16 by ngxson

server: wrap headers for mcp proxy examples server

#21072 opened 2026-03-27 15:18 by ngxson

hexagon: optimize HMX matmul operations ggml Hexagon

#21071 opened 2026-03-27 15:05 by chraac

Add quantization recipes from custom recipe files testing examples

#21070 opened 2026-03-27 14:52 by bartowski1182

ggml: allow prefetching tensor overrides Nvidia GPU examples ggml SYCL Ascend NPU OpenCL IBM zDNN OpenVINO WebGPU

#21067 opened 2026-03-27 14:02 by am17an

[HIP] Bump ROCm version to 7.2.1 devops

#21066 opened 2026-03-27 13:30 by slojosic-amd

ggml : use 64 bytes aligned tile buffers ggml

#21058 opened 2026-03-27 07:45 by angt

ggml webgpu: update Vulkan backend CI to use self-hosted runner devops ggml WebGPU

#21052 opened 2026-03-27 02:58 by reeselevine

Add the tests that we want to run on external CI script python devops

#21051 opened 2026-03-27 01:46 by shreyajn

ggml webgpu: move quantized buffers to u32 types and some other changes for wider browser/device support ggml WebGPU

#21046 opened 2026-03-26 22:25 by reeselevine

model: add Falcon OCR support model examples python ggml Apple Metal

#21045 opened 2026-03-26 22:07 by avirajBevli

llama : rotate activations for better quantization

#21038 opened 2026-03-26 18:14 by ggerganov

webui: Add option to pre-encode conversation for faster next turns examples server

#21034 opened 2026-03-26 16:08 by allozaur

model: add support for nvidia/gpt-oss-puzzle-88B model python

#21032 opened 2026-03-26 15:38 by smpurkis

vulkan: add FA dequant for q4_1, q5_0, q5_1, iq4_nl Vulkan ggml

#21029 opened 2026-03-26 15:03 by mkoker

Vulkan Q4_0 Repack PoC Vulkan ggml

#21024 opened 2026-03-26 12:03 by 0cc4m