You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Starting with a minimal Debian or Ubuntu base image is recommended, as CUDA
189
+
images include many pre-installed components that increase container size.
190
+
While the relationship isn't strictly linear, larger container sizes generally
191
+
lead to longer cold-starts and build times. Begin with a lean base image and
192
+
add only essential components as needed.
193
+
</Tip>
193
194
194
-
#### Ubuntu-based CUDA Images
195
+
#### Public Docker Hub Images with Namespaces
195
196
196
-
All Ubuntu-based `nvidia/cuda`images that include Ubuntu are supported. These include the CUDA libraries necessary to provide GPU acceleration:
197
+
When using public Docker Hub images that include a namespace (e.g., `bob/infinity`, `huggingface/transformers`), you need to be logged in to Docker Hub locally, even though the image is public. This is because Cerebrium reads your `~/.docker/config.json`to authenticate image pulls.
Starting with a minimal Debian or Ubuntu base image is recommended, as CUDA
228
-
images include many pre-installed components that increase container size.
229
-
While the relationship isn't strictly linear, larger container sizes generally
230
-
lead to longer cold-starts and build times. Begin with a lean base image and
231
-
add only essential components as needed.
232
-
</Tip>
233
+
However, **private ECR images** require authentication. See [Using Private Docker Registries](/cerebrium/container-images/private-docker-registry) for setup instructions.
0 commit comments