The SambaNova DataScale is a fully-integrated hardware-software AI system for model training, fine-tuning, and inference.
Connection to a SambaNova node is a two-step process. The first step is to ssh to the login node. The second step is to log in to a SambaNova node from the login node.
Log in to the SambaNova login node from your local machine. This uses the MobilePASS+ token generated every time you log in to the system.
In the example below, replace ALCFUserID with your ALCF user id.
ssh [email protected]
Password: < MobilePASS+ code >Once you are on the login node, ssh to one of the SambaNova nodes. For example:
ssh sn30-r1-h1 You can also ssh to sn30-r1-h2, sn30-r2-h1, sn30-r2-h2, sn30-r3-h1, sn30-r3-h2, sn30-r4-h1, sn30-r4-h2.
Use the examples of well-known AI applications under the path: /opt/sambaflow/apps, on all SambaNova compute nodes.
Make a copy of this to your home directory:
cp -r /opt/sambaflow/apps/ ~SambaNova software stack and associated environmental variables are setup at login.
Each of the samples or application examples provided by SambaNova has its own pre-built virtual environment which can be readily used. They are present in the /opt/sambaflow/apps/ directory tree within each of the applications.
Try additional examples.
- ALCF SambaNova documentation
- SambaNova documentation
- SambaNova applications path:
/opt/sambaflow/apps/ - SambaNova model scripts:
/data/ANL/scripts/ - Important datasets:
/software/sambanova/dataset/