Skip to content

Commit 3584e07

Browse files
localai-botmudlerlocalai-org-maint-bot
authored
chore: ⬆️ Update antirez/ds4 to efdadd41e20134af4f3381e1ed90e96fe4faef6f (#11010)
* ⬆️ Update antirez/ds4 Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> * fix(ds4): link new tensor parallel objects The updated ds4 revision split tensor-parallel transport and layer placement into separate translation units. Build and link those objects on CPU, CUDA, and Metal builds. Assisted-by: Codex:gpt-5 --------- Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com> Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>
1 parent 4ce67cc commit 3584e07

2 files changed

Lines changed: 16 additions & 16 deletions

File tree

backend/cpp/ds4/CMakeLists.txt

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -76,12 +76,13 @@ elseif(DS4_GPU STREQUAL "cpu")
7676
set(DS4_OBJS "${DS4_DIR}/ds4_cpu.o")
7777
endif()
7878

79-
# ds4.c now references ds4_distributed.c (distributed inference) and ds4_ssd.c
80-
# (SSD expert-cache), each split into its own translation unit upstream. Both
81-
# are GPU-agnostic objects shared by every GPU mode, so link them in regardless
82-
# of DS4_GPU.
79+
# Upstream splits distributed inference, tensor-parallel transport, the SSD
80+
# expert cache, and layer placement into GPU-agnostic translation units. Link
81+
# them regardless of DS4_GPU.
8382
list(APPEND DS4_OBJS "${DS4_DIR}/ds4_distributed.o")
83+
list(APPEND DS4_OBJS "${DS4_DIR}/ds4_tp.o")
8484
list(APPEND DS4_OBJS "${DS4_DIR}/ds4_ssd.o")
85+
list(APPEND DS4_OBJS "${DS4_DIR}/ds4_layer_pack.o")
8586

8687
add_executable(${TARGET}
8788
grpc-server.cpp

backend/cpp/ds4/Makefile

Lines changed: 11 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -1,10 +1,10 @@
11
# ds4 backend Makefile.
22
#
3-
# Upstream pin lives below as DS4_VERSION?=80ebbc396aee40eedc1d829222f3362d10fa4c6c
3+
# Upstream pin lives below as DS4_VERSION?=efdadd41e20134af4f3381e1ed90e96fe4faef6f
44
# (.github/bump_deps.sh) can find and update it - matches the
55
# llama-cpp / ik-llama-cpp / turboquant convention.
66

7-
DS4_VERSION?=80ebbc396aee40eedc1d829222f3362d10fa4c6c
7+
DS4_VERSION?=efdadd41e20134af4f3381e1ed90e96fe4faef6f
88
DS4_REPO?=https://github.com/antirez/ds4
99

1010
CURRENT_MAKEFILE_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
@@ -18,20 +18,19 @@ UNAME_S := $(shell uname -s)
1818

1919
CMAKE_ARGS ?= -DCMAKE_BUILD_TYPE=Release
2020

21-
# ds4_distributed.o and ds4_ssd.o are GPU-agnostic translation units that
22-
# ds4.c/ds4_cpu.o now reference (upstream split distributed inference and the
23-
# SSD expert-cache into their own .c files). Both objects are shared by every
24-
# GPU mode, so they are appended unconditionally below.
21+
# Upstream splits distributed inference, tensor-parallel transport, the SSD
22+
# expert cache, and layer placement into GPU-agnostic translation units. They
23+
# are shared by every GPU mode, so append them unconditionally below.
2524
ifeq ($(BUILD_TYPE),cublas)
2625
CMAKE_ARGS += -DDS4_GPU=cuda
27-
DS4_OBJ_TARGET := ds4.o ds4_cuda.o ds4_distributed.o ds4_ssd.o
26+
DS4_OBJ_TARGET := ds4.o ds4_cuda.o ds4_distributed.o ds4_tp.o ds4_ssd.o ds4_layer_pack.o
2827
else ifeq ($(UNAME_S),Darwin)
2928
CMAKE_ARGS += -DDS4_GPU=metal
30-
DS4_OBJ_TARGET := ds4.o ds4_metal.o ds4_distributed.o ds4_ssd.o
29+
DS4_OBJ_TARGET := ds4.o ds4_metal.o ds4_distributed.o ds4_tp.o ds4_ssd.o ds4_layer_pack.o
3130
else
3231
# CPU reference path (Linux only - macOS CPU path is broken by VM bug per ds4 README).
3332
CMAKE_ARGS += -DDS4_GPU=cpu
34-
DS4_OBJ_TARGET := ds4_cpu.o ds4_distributed.o ds4_ssd.o
33+
DS4_OBJ_TARGET := ds4_cpu.o ds4_distributed.o ds4_tp.o ds4_ssd.o ds4_layer_pack.o
3534
endif
3635

3736
ifneq ($(NATIVE),true)
@@ -56,11 +55,11 @@ ds4:
5655
# the right per-platform compile flags (Objective-C/Metal on Darwin, nvcc on Linux+CUDA).
5756
ds4/ds4.o: ds4
5857
ifeq ($(BUILD_TYPE),cublas)
59-
+$(MAKE) -C ds4 ds4.o ds4_cuda.o ds4_distributed.o ds4_ssd.o
58+
+$(MAKE) -C ds4 ds4.o ds4_cuda.o ds4_distributed.o ds4_tp.o ds4_ssd.o ds4_layer_pack.o
6059
else ifeq ($(UNAME_S),Darwin)
61-
+$(MAKE) -C ds4 ds4.o ds4_metal.o ds4_distributed.o ds4_ssd.o
60+
+$(MAKE) -C ds4 ds4.o ds4_metal.o ds4_distributed.o ds4_tp.o ds4_ssd.o ds4_layer_pack.o
6261
else
63-
+$(MAKE) -C ds4 ds4_cpu.o ds4_distributed.o ds4_ssd.o
62+
+$(MAKE) -C ds4 ds4_cpu.o ds4_distributed.o ds4_tp.o ds4_ssd.o ds4_layer_pack.o
6463
endif
6564

6665
grpc-server: ds4/ds4.o

0 commit comments

Comments
 (0)