Hi, My algorithm requires shm-size=8g. When debugging in docker container, I can easily set shm-size=8g this parameter, but I am not familiar with MLCude. How can I set it here?
Created by yuwan wang wanziya Hi Yuwan,
We have contacted you at your email address (wanziya@synapse.org) regarding your question and submission files. Could you please confirm if you have received our email by replying to the email thread?
Thank you very much.
Kind Regards,
LISA 2024 Challenge Organizers Could you also try running it without setting that option in your MLCube configuration file (MLCube.yaml)? We will email you to ask for more information that might help provide better guidance.
Kind Regards,
LISA 2024 Challenge Organizers Hi Yuwan,
Thank you for your question. To set shm-size=8g in your MLCube configuration file, please try to add it to the docker.run_args section of your mlcube.yaml file. This should allow you to specify additional Docker run options, such as configuring the shared memory size.
docker:
image: myimage:latest
run_args:
- "--shm-size=8g"
You can also set it directly in your docker run command by adding the argument --shm-size=8g. Hopefully, this resolves the issue. Please let us know if it works or not.
Best regards,
LISA 2024 Challenge Organizers It still happened.
```Bus error (core dumped)```.
I think this method may not work, because i see the command context from MLcube do not change or add ```shm-size=8g``` when i modify run_args in MLcube.
```context: {'status': 'signalled', 'code': -2, 'cmd': 'docker run --gpus=all --entrypoint=sh ```
But i directly set ```shm-size=8g``` in docker command, it will work successfully.
Besides, this is my config, maybe something error here i think.
```
platform:
accelerator_count: 0
docker:
image: docker.synapse.org/synxxx/lisa:v1
run_args:
- "--shm-size=8g"
tasks:
test:
parameters:
inputs:
data: { type: directory, default: "data" }
outputs:
results: { type: directory, default: "results" }
entrypoint: sh predict.sh
workspace:
# Local directory for inputs.
host: /App_mlcube
container: /workspace
```