llama.cpp
llama: add default load-mode auto, which avoids mmap on iGPUs
#26081

Merged

llama: add default load-mode auto, which avoids mmap on iGPUs #26081

ggerganov merged 9 commits into master from 0cc4m/mmap-auto

0cc4m requested a review from

CISC 37 days ago

0cc4m requested a review from

ggerganov 37 days ago

0cc4m requested a review 37 days ago

0cc4m requested a review from

marty1885 37 days ago

0cc4m requested a review 37 days ago

0cc4m requested a review from

JohannesGaessler 37 days ago

0cc4m requested a review 37 days ago

github-actions added Vulkan

github-actions added ggml

github-actions added SYCL

github-actions added Apple Metal

github-actions added Ascend NPU

github-actions added OpenCL

github-actions added IBM zDNN

github-actions added Hexagon

github-actions added CUDA

github-actions added AMD ZenDNN

github-actions added OpenVINO

github-actions added WebGPU

pwilkin approved these changes on 2026-07-24

ngxson approved these changes on 2026-07-24

taronaeo approved these changes on 2026-07-24

max-krasnyansky commented on 2026-07-24

max-krasnyansky approved these changes on 2026-07-24

taronaeo commented on 2026-07-24

lhez approved these changes on 2026-07-25

ORippler commented on 2026-07-27

ORippler assigned

ORippler 34 days ago

github-actions added examples

0cc4m force pushed to 36b9da05 33 days ago

woodpenker commented on 2026-07-29

ORippler commented on 2026-07-29

llama: add new default load-mode auto which picks mmap unless a non-M…

01649e61

Update ggml/src/ggml-hexagon/ggml-hexagon.cpp

9d8bd143

set mmap_support to false on OpenCL backend

4b4fada1

fix order of load modes

2f2d2465

use -1 for auto

9391a09f

resolve load mode auto earlier to correctly pick gpu host or cpu memory

18c23998

add load mode auto to llama-bench

f00aaaa6

bump virtgpu api version, regenerate docs

125b6c6f

0cc4m force pushed from 36b9da05 to 125b6c6f 25 days ago

0cc4m requested a review 25 days ago

github-actions added documentation

github-actions added server

ORippler approved these changes on 2026-08-05

0cc4m added merge ready

Merge branch 'master' into 0cc4m/mmap-auto

233ec61d

ggerganov merged 153d324b into master 19 days ago

ggerganov deleted the 0cc4m/mmap-auto branch 19 days ago

Reviewers

ORippler

lhez

taronaeo

max-krasnyansky

ngxson

pwilkin

woodpenker

CISC

ggerganov

marty1885

JohannesGaessler

Assignees

ORippler

Labels

documentation Vulkan examples server ggml merge ready SYCL Apple Metal Ascend NPU OpenCL IBM zDNN Hexagon CUDA AMD ZenDNN OpenVINO WebGPU

Milestone

No milestone

llama.cpp llama: add default load-mode auto, which avoids mmap on iGPUs #26081 Merged

llama: add default load-mode auto, which avoids mmap on iGPUs #26081

llama.cpp
llama: add default load-mode auto, which avoids mmap on iGPUs
#26081

Merged