Connection to one of the CS-2 cluster login nodes requires an MFA passcode for authentication - either an 8-digit passcode generated by an app on your mobile device (e.g. MobilePASS+) or a CRYPTOCard-generated passcode prefixed by a 4-digit pin. This is the same passcode used to authenticate into other ALCF systems, such as Theta and Cooley.
To connect to a CS-2 login, ssh to login nodes:
#Make your home directory navigable
chmod a+xr ~/
mkdir ~/R_1.9.2
chmod a+x ~/R_1.9.2/
cd ~/R_1.9.2
# Note: "deactivate" does not actually work in scripts.
deactivate
rm -r venv_pt
/software/cerebras/python3.8/bin/python3.8 -m venv venv_pt
source venv_pt/bin/activate
pip3 install /opt/cerebras/wheels/cerebras_pytorch-1.9.2+92b4fad15b-cp38-cp38-linux_x86_64.whl --find-links=/opt/cerebras/wheels
pip install numpy==1.23.4
pip install datasets transformers
We use example from Cerebras Modelzoo repository for this hands-on.
Clone the modezoo repository.
Note: For virtual environent R_1.9.2, the modelzoo is unchanged from R_1.9.1.
mkdir ~/R_1.9.2
cd ~/R_1.9.2
git clone https://github.com/Cerebras/modelzoo.git
cd modelzoo
git tag
git checkout Release_1.9.1
The CS-2 cluster has its own Kubernetes-based system for job submission and queuing. Jobs are started automatically through the Python scripts.
Use Cerebras cluster command line tool to get addional information about the jobs.
- Jobs that have not yet completed can be listed as
(venv_pt) $ csctl get jobs
- Jobs can be canceled as shown:
(venv_tf) $ csctl cancel job wsjob-eyjapwgnycahq9tus4w7id
See csctl -h
for more options.
Refer to respective instrcutions below
- ALCF Cerebras Documentation
- Cerebras Documntation
- Cerebras Modelzoo Repo
- Datasets Path:
/software/cerebras/dataset