Work on the `graph_optimize` method#23

Open

kpouget wants to merge 29 commits intocrc-org:mainfrom

kpouget:rebase-b7356-optimize

Commits on Dec 11, 2025

OWNERS: add file for OpenShift CI control
kpouget
committed
Add helper scripts
kpouget
committed
ggml: add the ggml-remotingfrontend and ggml-remotingbackend libraries
kpouget
committed
ggml: src: ggml-remotingfrontend/ggml-backend: add stub for .graph_optimize
kpouget
committed
src: llama-*: reduce the verbosity
kpouget
committed
tools: run: run: add timing instrumentation
kpouget
committed
ggml-metal: make less verbose
kpouget
committed
HACK: ggml-cpu: reduce the verbosity
kpouget
committed
ggml-remotingbackend: add missing includes
kpouget
committed
ggml/src/ggml-backend-reg.cpp: fix the frontend library name
kpouget
committed
ggml-remotingbackend/backend: disable the GGML_BACKEND_LIBRARY_METAL_DEVICE_CONTEXT lookup
kpouget
committed
ggml-remotingfrontend: disable USE_METAL_GUEST_SUPPORTS_OP
kpouget
committed
ggml-remotingfrontend/ggml-backend-reg: don't initialize the metal context when disabled
kpouget
committed
ggml-remotingfrontend: disable USE_FROM_PTR (for Linux)
kpouget
committed
ggml: src: ggml-remotingfrontend/ggml-remoting: use the arch to distinguish linux vs macos
kpouget
committed
Update to make it work with Linux
kpouget
committed
ggml: src: ggml-remotingfrontend/ggml-backend-buffer: don't stop on buffer_cpy_tensor(
kpouget
committed
ggml: force disable vulkan loader when compiling the ggml-remotingbackend
kpouget
committed
ggml-remoting: add support for the buffer_cpy_tensor function
kpouget
committed
ggml-remotingbackend: cleanups
kpouget
committed
ggml-remotingbackend: fix inconsist fatal message ...
kpouget
committed
ggml-remotingfrontend: properly distinguish if the backend supports buffer_from_host_ptr
kpouget
committed
Revert "metal : make the FA extra sizes consistent (#17143)"
kpouget
committed
ggml-remoting: add the code the the graph_optimize function
kpouget
committed
ggml-remoting: update the graph serial/deserial
kpouget
committed
Start integrating code for returning the optimized cgraph
kpouget
committed
Add code to track the frontend buffers
kpouget
committed
ggml-remotingfrontend: lookup the guest buffer from the host handle
kpouget
committed
HACK: ggml-backend-reg: allow disabling the Vulkan backend
kpouget
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Work on the `graph_optimize` method#23

Work on the `graph_optimize` method#23
kpouget wants to merge 29 commits intocrc-org:mainfrom
kpouget:rebase-b7356-optimize

Commits on Dec 11, 2025

OWNERS: add file for OpenShift CI control

Add helper scripts

ggml: add the ggml-remotingfrontend and ggml-remotingbackend libraries

ggml: src: ggml-remotingfrontend/ggml-backend: add stub for .graph_optimize

src: llama-*: reduce the verbosity

tools: run: run: add timing instrumentation

ggml-metal: make less verbose

HACK: ggml-cpu: reduce the verbosity

ggml-remotingbackend: add missing includes

ggml/src/ggml-backend-reg.cpp: fix the frontend library name

ggml-remotingbackend/backend: disable the GGML_BACKEND_LIBRARY_METAL_DEVICE_CONTEXT lookup

ggml-remotingfrontend: disable USE_METAL_GUEST_SUPPORTS_OP

ggml-remotingfrontend/ggml-backend-reg: don't initialize the metal context when disabled

ggml-remotingfrontend: disable USE_FROM_PTR (for Linux)

ggml: src: ggml-remotingfrontend/ggml-remoting: use the arch to distinguish linux vs macos

Update to make it work with Linux

ggml: src: ggml-remotingfrontend/ggml-backend-buffer: don't stop on buffer_cpy_tensor(

ggml: force disable vulkan loader when compiling the ggml-remotingbackend

ggml-remoting: add support for the buffer_cpy_tensor function

ggml-remotingbackend: cleanups

ggml-remotingbackend: fix inconsist fatal message ...

ggml-remotingfrontend: properly distinguish if the backend supports buffer_from_host_ptr

Revert "metal : make the FA extra sizes consistent (#17143)"

ggml-remoting: add the code the the graph_optimize function

ggml-remoting: update the graph serial/deserial

Start integrating code for returning the optimized cgraph

Add code to track the frontend buffers

ggml-remotingfrontend: lookup the guest buffer from the host handle

HACK: ggml-backend-reg: allow disabling the Vulkan backend