Skip to content

Commit d9e25af

Browse files
mudlermudler
authored
feat(nvidia-gpu): bump images to cuda 12.8 (#6239)
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
1 parent e4f8b63 commit d9e25af

File tree

6 files changed

+25
-25
lines changed

6 files changed

+25
-25
lines changed

‎.github/workflows/backend.yml‎

Lines changed: 19 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -89,7 +89,7 @@ jobs:
8989
context: "./backend"
9090
- build-type: 'l4t'
9191
cuda-major-version: "12"
92-
cuda-minor-version: "0"
92+
cuda-minor-version: "8"
9393
platforms: 'linux/arm64'
9494
tag-latest: 'auto'
9595
tag-suffix: '-gpu-nvidia-l4t-diffusers'
@@ -175,7 +175,7 @@ jobs:
175175
# CUDA 12 builds
176176
- build-type: 'cublas'
177177
cuda-major-version: "12"
178-
cuda-minor-version: "0"
178+
cuda-minor-version: "8"
179179
platforms: 'linux/amd64'
180180
tag-latest: 'auto'
181181
tag-suffix: '-gpu-nvidia-cuda-12-rerankers'
@@ -187,7 +187,7 @@ jobs:
187187
context: "./backend"
188188
- build-type: 'cublas'
189189
cuda-major-version: "12"
190-
cuda-minor-version: "0"
190+
cuda-minor-version: "8"
191191
platforms: 'linux/amd64'
192192
tag-latest: 'auto'
193193
tag-suffix: '-gpu-nvidia-cuda-12-llama-cpp'
@@ -199,7 +199,7 @@ jobs:
199199
context: "./"
200200
- build-type: 'cublas'
201201
cuda-major-version: "12"
202-
cuda-minor-version: "0"
202+
cuda-minor-version: "8"
203203
platforms: 'linux/amd64'
204204
tag-latest: 'auto'
205205
tag-suffix: '-gpu-nvidia-cuda-12-vllm'
@@ -211,7 +211,7 @@ jobs:
211211
context: "./backend"
212212
- build-type: 'cublas'
213213
cuda-major-version: "12"
214-
cuda-minor-version: "0"
214+
cuda-minor-version: "8"
215215
platforms: 'linux/amd64'
216216
tag-latest: 'auto'
217217
tag-suffix: '-gpu-nvidia-cuda-12-transformers'
@@ -223,7 +223,7 @@ jobs:
223223
context: "./backend"
224224
- build-type: 'cublas'
225225
cuda-major-version: "12"
226-
cuda-minor-version: "0"
226+
cuda-minor-version: "8"
227227
platforms: 'linux/amd64'
228228
tag-latest: 'auto'
229229
tag-suffix: '-gpu-nvidia-cuda-12-diffusers'
@@ -236,7 +236,7 @@ jobs:
236236
# CUDA 12 additional backends
237237
- build-type: 'cublas'
238238
cuda-major-version: "12"
239-
cuda-minor-version: "0"
239+
cuda-minor-version: "8"
240240
platforms: 'linux/amd64'
241241
tag-latest: 'auto'
242242
tag-suffix: '-gpu-nvidia-cuda-12-kokoro'
@@ -248,7 +248,7 @@ jobs:
248248
context: "./backend"
249249
- build-type: 'cublas'
250250
cuda-major-version: "12"
251-
cuda-minor-version: "0"
251+
cuda-minor-version: "8"
252252
platforms: 'linux/amd64'
253253
tag-latest: 'auto'
254254
tag-suffix: '-gpu-nvidia-cuda-12-faster-whisper'
@@ -260,7 +260,7 @@ jobs:
260260
context: "./backend"
261261
- build-type: 'cublas'
262262
cuda-major-version: "12"
263-
cuda-minor-version: "0"
263+
cuda-minor-version: "8"
264264
platforms: 'linux/amd64'
265265
tag-latest: 'auto'
266266
tag-suffix: '-gpu-nvidia-cuda-12-coqui'
@@ -272,7 +272,7 @@ jobs:
272272
context: "./backend"
273273
- build-type: 'cublas'
274274
cuda-major-version: "12"
275-
cuda-minor-version: "0"
275+
cuda-minor-version: "8"
276276
platforms: 'linux/amd64'
277277
tag-latest: 'auto'
278278
tag-suffix: '-gpu-nvidia-cuda-12-bark'
@@ -284,7 +284,7 @@ jobs:
284284
context: "./backend"
285285
- build-type: 'cublas'
286286
cuda-major-version: "12"
287-
cuda-minor-version: "0"
287+
cuda-minor-version: "8"
288288
platforms: 'linux/amd64'
289289
tag-latest: 'auto'
290290
tag-suffix: '-gpu-nvidia-cuda-12-chatterbox'
@@ -566,7 +566,7 @@ jobs:
566566
context: "./"
567567
- build-type: 'cublas'
568568
cuda-major-version: "12"
569-
cuda-minor-version: "0"
569+
cuda-minor-version: "8"
570570
platforms: 'linux/arm64'
571571
skip-drivers: 'true'
572572
tag-latest: 'auto'
@@ -603,7 +603,7 @@ jobs:
603603
context: "./"
604604
- build-type: 'cublas'
605605
cuda-major-version: "12"
606-
cuda-minor-version: "0"
606+
cuda-minor-version: "8"
607607
platforms: 'linux/amd64'
608608
tag-latest: 'auto'
609609
tag-suffix: '-gpu-nvidia-cuda-12-stablediffusion-ggml'
@@ -663,7 +663,7 @@ jobs:
663663
context: "./"
664664
- build-type: 'cublas'
665665
cuda-major-version: "12"
666-
cuda-minor-version: "0"
666+
cuda-minor-version: "8"
667667
platforms: 'linux/arm64'
668668
skip-drivers: 'true'
669669
tag-latest: 'auto'
@@ -688,7 +688,7 @@ jobs:
688688
context: "./"
689689
- build-type: 'cublas'
690690
cuda-major-version: "12"
691-
cuda-minor-version: "0"
691+
cuda-minor-version: "8"
692692
platforms: 'linux/amd64'
693693
tag-latest: 'auto'
694694
tag-suffix: '-gpu-nvidia-cuda-12-whisper'
@@ -748,7 +748,7 @@ jobs:
748748
context: "./"
749749
- build-type: 'cublas'
750750
cuda-major-version: "12"
751-
cuda-minor-version: "0"
751+
cuda-minor-version: "8"
752752
platforms: 'linux/arm64'
753753
skip-drivers: 'true'
754754
tag-latest: 'auto'
@@ -824,7 +824,7 @@ jobs:
824824
context: "./backend"
825825
- build-type: 'cublas'
826826
cuda-major-version: "12"
827-
cuda-minor-version: "0"
827+
cuda-minor-version: "8"
828828
platforms: 'linux/amd64'
829829
tag-latest: 'auto'
830830
tag-suffix: '-gpu-nvidia-cuda-12-rfdetr'
@@ -860,7 +860,7 @@ jobs:
860860
context: "./backend"
861861
- build-type: 'cublas'
862862
cuda-major-version: "12"
863-
cuda-minor-version: "0"
863+
cuda-minor-version: "8"
864864
platforms: 'linux/arm64'
865865
skip-drivers: 'true'
866866
tag-latest: 'auto'
@@ -885,7 +885,7 @@ jobs:
885885
context: "./backend"
886886
- build-type: 'cublas'
887887
cuda-major-version: "12"
888-
cuda-minor-version: "0"
888+
cuda-minor-version: "8"
889889
platforms: 'linux/amd64'
890890
tag-latest: 'auto'
891891
tag-suffix: '-gpu-nvidia-cuda-12-exllama2'

‎.github/workflows/image-pr.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -36,7 +36,7 @@ jobs:
3636
include:
3737
- build-type: 'cublas'
3838
cuda-major-version: "12"
39-
cuda-minor-version: "0"
39+
cuda-minor-version: "8"
4040
platforms: 'linux/amd64'
4141
tag-latest: 'false'
4242
tag-suffix: '-gpu-nvidia-cuda-12'

‎.github/workflows/image.yml‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -91,7 +91,7 @@ jobs:
9191
aio: "-aio-gpu-nvidia-cuda-11"
9292
- build-type: 'cublas'
9393
cuda-major-version: "12"
94-
cuda-minor-version: "0"
94+
cuda-minor-version: "8"
9595
platforms: 'linux/amd64'
9696
tag-latest: 'auto'
9797
tag-suffix: '-gpu-nvidia-cuda-12'
@@ -144,7 +144,7 @@ jobs:
144144
include:
145145
- build-type: 'cublas'
146146
cuda-major-version: "12"
147-
cuda-minor-version: "0"
147+
cuda-minor-version: "8"
148148
platforms: 'linux/arm64'
149149
tag-latest: 'auto'
150150
tag-suffix: '-nvidia-l4t-arm64'

‎Dockerfile‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -18,7 +18,7 @@ FROM requirements AS requirements-drivers
1818

1919
ARG BUILD_TYPE
2020
ARG CUDA_MAJOR_VERSION=12
21-
ARG CUDA_MINOR_VERSION=0
21+
ARG CUDA_MINOR_VERSION=8
2222
ARG SKIP_DRIVERS=false
2323
ARG TARGETARCH
2424
ARG TARGETVARIANT

‎Makefile‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -170,7 +170,7 @@ prepare-e2e:
170170
mkdir -p $(TEST_DIR)
171171
cp -rfv $(abspath ./tests/e2e-fixtures)/gpu.yaml $(TEST_DIR)/gpu.yaml
172172
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 $(TEST_DIR)/ggllm-test-model.bin
173-
docker build --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=0 -t localai-tests .
173+
docker build --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=8 -t localai-tests .
174174

175175
run-e2e-image:
176176
ls -liah $(abspath ./tests/e2e-fixtures)

‎backend/README.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -111,7 +111,7 @@ docker build -f backend/Dockerfile.python \
111111
--build-arg BACKEND=transformers \
112112
--build-arg BUILD_TYPE=cublas12 \
113113
--build-arg CUDA_MAJOR_VERSION=12 \
114-
--build-arg CUDA_MINOR_VERSION=0 \
114+
--build-arg CUDA_MINOR_VERSION=8 \
115115
-t localai-backend-transformers .
116116

117117
# Build Go backend

0 commit comments

Comments
 (0)