Groq
Getting started with Groq.
Last updated
Was this helpful?
Getting started with Groq.
Last updated
Was this helpful?
Start by logging into ExCL's login node.
From the login node, you can then login to a node with a Groq card, for example
Here is a table of the Groq cards available:
milan1
1
milan2
1
The recommended way to access the Groq card is to reserve it through the Slurm resource manager. Groq cards are available on machines in the groq partition. To reserve a node with a groq card for interactive use use the command.
Where:
-J
,--job-name=<jobname>
specifies the job name.-p
,--partition=<partition names>
specifies the partition name.--exclusive
specifies you want exclusive access to the node.--gres="groq:card:1"
specifies that you want to use 1 groq card.
Non-interactive batch jobs can similarly be launched.
Where:
-J
,--job-name=<jobname>
specifies the job name.-p
,--partition=<partition names>
specifies the partition name.--exclusive
specifies you want exclusive access to the node.--gres="groq:card:1"
specifies that you want to use 1 groq card.
or specified in the script:
In order to use the Groq API you need to make sure you are using python 3.8 and that you add the Groq python libraries to your path. For python 3.8 you can either use the installed system python3.8 or use conda to install python3.8.
You need to fully quantify the path to python since Ubuntu 22.04 defaults to python3.10. This means you need to use
Then to install jupyter notebook in your home directory, you would need to do
Run regression tests to verify card functionality: /opt/groq/runtime/site-packages/bin/tsp-regression run
Get Groq device status: /opt/groq/runtime/site-packages/bin/tsp-ctl status
Monitor temperature and power: /opt/groq/runtime/site-packages/bin/tsp-ctl monitor
First install miniconda by following . Then create a groq environment with
See the for more details for setting up the Conda environment.
See .
See for more information on setting up Jupyter Notebooks within ExCL.