feat: support ggml backend select and enhance the build system#1368

Open

Cyberhan123 wants to merge 32 commits intoleejet:masterfrom

Cyberhan123:ggml-new-backend

Commits on Mar 26, 2026

Select backend devices via arg

stduhpf
authored and
Cyberhan123
committed
fix build

stduhpf
authored and
Cyberhan123
committed
show backend device description

stduhpf
authored and
Cyberhan123
committed
CLI: add --list-devices arg

stduhpf
authored and
Cyberhan123
committed
null-terminate even if buffer is too small

stduhpf
authored and
Cyberhan123
committed
move stuff to ggml_extend.cpp

stduhpf
authored and
Cyberhan123
committed
--upscaler-backend-device

stduhpf
authored and
Cyberhan123
committed
use diffusion_backend for loading LoRAs

stduhpf
authored and
Cyberhan123
committed
--photomaker-backend-device (+fixes)

stduhpf
authored and
Cyberhan123
committed
--vision-backend-device

stduhpf
authored and
Cyberhan123
committed
check backends at runtime

Stéphane du Hamel
authored and
Cyberhan123
committed
fix missing includes

Stéphane du Hamel
authored and
Cyberhan123
committed
fix typo

Stéphane du Hamel
authored and
Cyberhan123
committed
multiple clip backend devices

Stéphane du Hamel
authored and
Cyberhan123
committed
update help message

stduhpf
authored and
Cyberhan123
committed
Add RPC documentation

stduhpf
authored and
Cyberhan123
committed
update docs

stduhpf
authored and
Cyberhan123
committed
update RPC docs

stduhpf
authored and
Cyberhan123
committed
fix apply_loras_immediately when using different non-CPU backends

stduhpf
authored and
Cyberhan123
committed
Force sequencial tensor loading when using RPC

stduhpf
authored and
Cyberhan123
committed
fix build

stduhpf
authored and
Cyberhan123
committed
Get first stage backend for loading loras

stduhpf
authored and
Cyberhan123
committed
Fix lora loading when using multiple clip backends

stduhpf
authored and
Cyberhan123
committed
rm all head code for ggml
Cyberhan123
committed
feat: enhance the build system by adding CMake configuration and backend support, and eliminate hard-coded backend references.
Cyberhan123
committed
fix review
Cyberhan123
committed
format code
Cyberhan123
committed
up date ci job
Cyberhan123
committed
fix vision_backend null
Cyberhan123
committed
rm -j args
Cyberhan123
committed
test ci
Cyberhan123
committed
fix: remove trailing backslashes in CMake commands
Cyberhan123
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: support ggml backend select and enhance the build system#1368

feat: support ggml backend select and enhance the build system#1368
Cyberhan123 wants to merge 32 commits intoleejet:masterfrom
Cyberhan123:ggml-new-backend

Commits on Mar 26, 2026

Select backend devices via arg

fix build

show backend device description

CLI: add --list-devices arg

null-terminate even if buffer is too small

move stuff to ggml_extend.cpp

--upscaler-backend-device

use diffusion_backend for loading LoRAs

--photomaker-backend-device (+fixes)

--vision-backend-device

check backends at runtime

fix missing includes

fix typo

multiple clip backend devices

update help message

Add RPC documentation

update docs

update RPC docs

fix apply_loras_immediately when using different non-CPU backends

Force sequencial tensor loading when using RPC

fix build

Get first stage backend for loading loras

Fix lora loading when using multiple clip backends

rm all head code for ggml

feat: enhance the build system by adding CMake configuration and backend support, and eliminate hard-coded backend references.

fix review

format code

up date ci job

fix vision_backend null

rm -j args

test ci

fix: remove trailing backslashes in CMake commands