What to use it for
Cerebras systems are designed for large-scale high-throughput training, tuning and inference of deep learning tasks such as Large Language Models. The system has libraries for Python (extension of pytorch) and Cerebras offers a model zoo to get going quickly.
Service Profile
The EIDF Cerebras Service is comprised of 4 CS3 Wafer-Scale Engines connected via a SwarmX switch to form a Wafer-Scale Cluster. The addition of a MemoryX unit allows streaming of data into and out of each unit allowing training of models which exceed the memory capacity of a single unit.
Availability
This service is generally as part of EIDF. Please contact the EIDF service desk for information on how to apply for access.