Add documentation for making custom min container #3059
Conversation
docs/compose.md
Outdated
| && apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub \ | ||
| && add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /" | ||
| RUN apt-get update \ | ||
| && apt-get install -y datacenter-gpu-manager |
There was a problem hiding this comment.
Fix indents here and above
There was a problem hiding this comment.
did you want me to indent everything? Not sure if this matters?
| Then use Docker to create the image. | ||
|
|
||
| ``` | ||
| $ docker build -t tritonserver_min . | ||
| $ docker build -t tritonserver_min -f <Dockerfile name> . |
There was a problem hiding this comment.
Dockerbuild assumes default dockerfile name to be PATH/Dockerfile so we want to specify here
docs/compose.md
Outdated
| @@ -67,11 +67,33 @@ FROM nvcr.io/nvidia/tritonserver:<xx.yy>-py3-min | |||
| COPY --from=full /opt/tritonserver/bin /opt/tritonserver/bin | |||
| COPY --from=full /opt/tritonserver/lib /opt/tritonserver/lib | |||
| ``` | |||
| Then install dependencies outlined in [build.py](https://github.com/triton-inference-server/server/blob/main/build.py#L670-L694) | |||
There was a problem hiding this comment.
Refer to function name instead of line numbers since the line numbers will quickly become out of date.
docs/compose.md
Outdated
|
|
||
| Then install dependencies outlined in [build.py](https://github.com/triton-inference-server/server/blob/main/build.py), `dockerfile_add_installation_linux()` function. | ||
|
|
||
| Example Dockerfile that only uses `Tensorflow 1` with the 21.06 build: |
There was a problem hiding this comment.
Example Dockerfile that creates a Triton image that contains only the TensorFlow 1 backend.
docs/compose.md
Outdated
| ``` | ||
| FROM nvcr.io/nvidia/tritonserver:21.06-py3 as full | ||
| FROM nvcr.io/nvidia/tritonserver:21.06-py3-min | ||
| COPY --from=full /opt/tritonserver/bin /opt/tritonserver/bin |
There was a problem hiding this comment.
I don't like this because it is not complete. I think we should stop documenting the manual way of doing compose and instead just change this documentation to only refer to the compose.py script and how to use it.
|
I wasn't clear in my earlier comment. I would not make any of these changes at all. As part of your compose.py change you should update the documentation so that it only talks about using compose.py and not about doing it yourself. One thing we can do is add a --dryrun flag to compose.py that stops after generating the dockerfile, so that anyone that wants to use that as a starting point to add additional changes can use that flag and then modify the generated dockerfile. |
|
all changes will be addressed in #3039 |
Addressing /github.com//issues/2910