Skip to content

Commit 43f0688

Browse files
reneleonhardtreneleonhardt
authored
feat: Upgrade to CUDA 12.5 (#2601)
Signed-off-by: Rene Leonhardt <[email protected]>
1 parent 8142bdc commit 43f0688

File tree

6 files changed

+42
-31
lines changed

6 files changed

+42
-31
lines changed

‎.github/workflows/image-pr.yml‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -46,7 +46,7 @@ jobs:
4646
makeflags: "--jobs=3 --output-sync=target"
4747
- build-type: 'cublas'
4848
cuda-major-version: "12"
49-
cuda-minor-version: "1"
49+
cuda-minor-version: "5"
5050
platforms: 'linux/amd64'
5151
tag-latest: 'false'
5252
tag-suffix: '-cublas-cuda12-ffmpeg'
@@ -119,12 +119,12 @@ jobs:
119119
makeflags: "--jobs=3 --output-sync=target"
120120
- build-type: 'cublas'
121121
cuda-major-version: "12"
122-
cuda-minor-version: "1"
122+
cuda-minor-version: "5"
123123
platforms: 'linux/amd64'
124124
tag-latest: 'false'
125125
tag-suffix: '-cublas-cuda12-ffmpeg-core'
126126
ffmpeg: 'true'
127127
image-type: 'core'
128128
runs-on: 'ubuntu-latest'
129129
base-image: "ubuntu:22.04"
130-
makeflags: "--jobs=4 --output-sync=target"
130+
makeflags: "--jobs=4 --output-sync=target"

‎.github/workflows/image.yml‎

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -64,7 +64,7 @@ jobs:
6464
makeflags: "--jobs=3 --output-sync=target"
6565
- build-type: 'cublas'
6666
cuda-major-version: "11"
67-
cuda-minor-version: "7"
67+
cuda-minor-version: "8"
6868
platforms: 'linux/amd64'
6969
tag-latest: 'false'
7070
tag-suffix: '-cublas-cuda11'
@@ -75,7 +75,7 @@ jobs:
7575
makeflags: "--jobs=3 --output-sync=target"
7676
- build-type: 'cublas'
7777
cuda-major-version: "12"
78-
cuda-minor-version: "1"
78+
cuda-minor-version: "5"
7979
platforms: 'linux/amd64'
8080
tag-latest: 'false'
8181
tag-suffix: '-cublas-cuda12'
@@ -86,7 +86,7 @@ jobs:
8686
makeflags: "--jobs=3 --output-sync=target"
8787
- build-type: 'cublas'
8888
cuda-major-version: "11"
89-
cuda-minor-version: "7"
89+
cuda-minor-version: "8"
9090
platforms: 'linux/amd64'
9191
tag-latest: 'auto'
9292
tag-suffix: '-cublas-cuda11-ffmpeg'
@@ -100,7 +100,7 @@ jobs:
100100
makeflags: "--jobs=3 --output-sync=target"
101101
- build-type: 'cublas'
102102
cuda-major-version: "12"
103-
cuda-minor-version: "1"
103+
cuda-minor-version: "5"
104104
platforms: 'linux/amd64'
105105
tag-latest: 'auto'
106106
tag-suffix: '-cublas-cuda12-ffmpeg'
@@ -232,7 +232,7 @@ jobs:
232232
grpc-base-image: "ubuntu:22.04"
233233
runs-on: 'arc-runner-set'
234234
makeflags: "--jobs=3 --output-sync=target"
235-
235+
236236
core-image-build:
237237
uses: ./.github/workflows/image_build.yml
238238
with:
@@ -273,7 +273,7 @@ jobs:
273273
makeflags: "--jobs=4 --output-sync=target"
274274
- build-type: 'cublas'
275275
cuda-major-version: "11"
276-
cuda-minor-version: "7"
276+
cuda-minor-version: "8"
277277
platforms: 'linux/amd64'
278278
tag-latest: 'false'
279279
tag-suffix: '-cublas-cuda11-core'
@@ -284,7 +284,7 @@ jobs:
284284
makeflags: "--jobs=4 --output-sync=target"
285285
- build-type: 'cublas'
286286
cuda-major-version: "12"
287-
cuda-minor-version: "1"
287+
cuda-minor-version: "5"
288288
platforms: 'linux/amd64'
289289
tag-latest: 'false'
290290
tag-suffix: '-cublas-cuda12-core'
@@ -295,7 +295,7 @@ jobs:
295295
makeflags: "--jobs=4 --output-sync=target"
296296
- build-type: 'cublas'
297297
cuda-major-version: "11"
298-
cuda-minor-version: "7"
298+
cuda-minor-version: "8"
299299
platforms: 'linux/amd64'
300300
tag-latest: 'false'
301301
tag-suffix: '-cublas-cuda11-ffmpeg-core'
@@ -306,7 +306,7 @@ jobs:
306306
makeflags: "--jobs=4 --output-sync=target"
307307
- build-type: 'cublas'
308308
cuda-major-version: "12"
309-
cuda-minor-version: "1"
309+
cuda-minor-version: "5"
310310
platforms: 'linux/amd64'
311311
tag-latest: 'false'
312312
tag-suffix: '-cublas-cuda12-ffmpeg-core'

‎.github/workflows/image_build.yml‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -19,11 +19,11 @@ on:
1919
type: string
2020
cuda-major-version:
2121
description: 'CUDA major version'
22-
default: "11"
22+
default: "12"
2323
type: string
2424
cuda-minor-version:
2525
description: 'CUDA minor version'
26-
default: "7"
26+
default: "5"
2727
type: string
2828
platforms:
2929
description: 'Platforms'
@@ -324,7 +324,7 @@ jobs:
324324
docker pull quay.io/go-skynet/local-ai:${{ steps.meta_aio.outputs.version }}
325325
docker tag quay.io/go-skynet/local-ai:${{ steps.meta_aio.outputs.version }} quay.io/go-skynet/local-ai:${{ inputs.latest-image-aio }}
326326
docker push quay.io/go-skynet/local-ai:${{ inputs.latest-image-aio }}
327-
327+
328328
- name: job summary
329329
run: |
330330
echo "Built image: ${{ steps.meta.outputs.labels }}" >> $GITHUB_STEP_SUMMARY

‎.github/workflows/release.yaml‎

Lines changed: 11 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@ jobs:
4040
sudo apt-get update
4141
sudo apt-get install -y cuda-cross-aarch64 cuda-nvcc-cross-aarch64-${CUDA_VERSION} libcublas-cross-aarch64-${CUDA_VERSION}
4242
env:
43-
CUDA_VERSION: 12-4
43+
CUDA_VERSION: 12-5
4444
- name: Cache grpc
4545
id: cache-grpc
4646
uses: actions/cache@v4
@@ -167,26 +167,26 @@ jobs:
167167
ROCM_VERSION: "6.1"
168168
AMDGPU_VERSION: "6.1"
169169
run: |
170-
set -ex
170+
set -ex
171171
172172
sudo apt-get update
173-
sudo DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends ca-certificates curl libnuma-dev gnupg
174-
175-
curl -sL https://repo.radeon.com/rocm/rocm.gpg.key | sudo apt-key add -
176-
173+
sudo DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends ca-certificates curl libnuma-dev gnupg
174+
175+
curl -sL https://repo.radeon.com/rocm/rocm.gpg.key | sudo apt-key add -
176+
177177
printf "deb [arch=amd64] https://repo.radeon.com/rocm/apt/$ROCM_VERSION/ jammy main" | sudo tee /etc/apt/sources.list.d/rocm.list
178-
178+
179179
printf "deb [arch=amd64] https://repo.radeon.com/amdgpu/$AMDGPU_VERSION/ubuntu jammy main" | sudo tee /etc/apt/sources.list.d/amdgpu.list
180180
printf 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600
181181
sudo apt-get update
182182
183183
sudo DEBIAN_FRONTEND=noninteractive apt-get install -y \
184184
hipblas-dev rocm-dev \
185185
rocblas-dev
186-
186+
187187
sudo apt-get clean
188188
sudo rm -rf /var/lib/apt/lists/*
189-
sudo ldconfig
189+
sudo ldconfig
190190
- name: Cache grpc
191191
id: cache-grpc
192192
uses: actions/cache@v4
@@ -291,7 +291,7 @@ jobs:
291291
export C_INCLUDE_PATH=/usr/local/include
292292
export CPLUS_INCLUDE_PATH=/usr/local/include
293293
export PATH=$PATH:$GOPATH/bin
294-
294+
295295
BACKEND_LIBS="$(ls /opt/homebrew/opt/grpc/lib/*.dylib /opt/homebrew/opt/re2/lib/*.dylib /opt/homebrew/opt/openssl@3/lib/*.dylib /opt/homebrew/opt/protobuf/lib/*.dylib /opt/homebrew/opt/abseil/lib/*.dylib | xargs)" GO_TAGS=p2p make dist
296296
- uses: actions/upload-artifact@v4
297297
with:
@@ -309,4 +309,4 @@ jobs:
309309
with:
310310
detached: true
311311
connect-timeout-seconds: 180
312-
limit-access-to-actor: true
312+
limit-access-to-actor: true

‎Dockerfile‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -33,7 +33,7 @@ RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | ta
3333
ENV PATH $PATH:/root/go/bin:/usr/local/go/bin
3434

3535
# Install grpc compilers
36-
RUN go install google.golang.org/protobuf/cmd/[email protected].1 && \
36+
RUN go install google.golang.org/protobuf/cmd/[email protected].2 && \
3737
go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
3838

3939
COPY --chmod=644 custom-ca-certs/* /usr/local/share/ca-certificates/
@@ -98,8 +98,8 @@ RUN pip install --user grpcio-tools
9898
FROM requirements-${IMAGE_TYPE} AS requirements-drivers
9999

100100
ARG BUILD_TYPE
101-
ARG CUDA_MAJOR_VERSION=11
102-
ARG CUDA_MINOR_VERSION=8
101+
ARG CUDA_MAJOR_VERSION=12
102+
ARG CUDA_MINOR_VERSION=5
103103

104104
ENV BUILD_TYPE=${BUILD_TYPE}
105105

@@ -292,7 +292,7 @@ ENV REBUILD=false
292292
ENV HEALTHCHECK_ENDPOINT=http://localhost:8080/readyz
293293
ENV MAKEFLAGS=${MAKEFLAGS}
294294

295-
ARG CUDA_MAJOR_VERSION=11
295+
ARG CUDA_MAJOR_VERSION=12
296296
ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
297297
ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
298298
ENV NVIDIA_VISIBLE_DEVICES=all

‎Makefile‎

Lines changed: 12 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -400,7 +400,7 @@ prepare-e2e:
400400
mkdir -p $(TEST_DIR)
401401
cp -rfv $(abspath ./tests/e2e-fixtures)/gpu.yaml $(TEST_DIR)/gpu.yaml
402402
test -e $(TEST_DIR)/ggllm-test-model.bin || wget -q https://huggingface.co/TheBloke/CodeLlama-7B-Instruct-GGUF/resolve/main/codellama-7b-instruct.Q2_K.gguf -O $(TEST_DIR)/ggllm-test-model.bin
403-
docker build --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=11 --build-arg CUDA_MINOR_VERSION=7 --build-arg FFMPEG=true -t localai-tests .
403+
docker build --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=5 --build-arg FFMPEG=true -t localai-tests .
404404

405405
run-e2e-image:
406406
ls -liah $(abspath ./tests/e2e-fixtures)
@@ -810,6 +810,17 @@ docker:
810810
--build-arg BUILD_TYPE=$(BUILD_TYPE) \
811811
-t $(DOCKER_IMAGE) .
812812

813+
docker-cuda11:
814+
docker build \
815+
--build-arg CUDA_MAJOR_VERSION=11 \
816+
--build-arg CUDA_MINOR_VERSION=8 \
817+
--build-arg BASE_IMAGE=$(BASE_IMAGE) \
818+
--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
819+
--build-arg GO_TAGS="$(GO_TAGS)" \
820+
--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
821+
--build-arg BUILD_TYPE=$(BUILD_TYPE) \
822+
-t $(DOCKER_IMAGE)-cuda11 .
823+
813824
docker-aio:
814825
@echo "Building AIO image with base $(BASE_IMAGE) as $(DOCKER_AIO_IMAGE)"
815826
docker build \

0 commit comments

Comments
 (0)