This is the document outlining the steps to run a Lagrange based ZKsync prover.
As the ZKsync prover makes heavy use of CUDA for fast computation, GPU passthru must be enabled in the Docker server to allow CUDA to be used from within a container.
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html
The Lagrange/ZKsync prover is distributed in a Docker image, available here:
For L4 GPU:
For A10 GPU:
Two sources of data will must be injected through mounts in the running docker image instance.
The configuration file must be edited according to the worker details. It will be mounted under /app/config/config.toml
.
Important: change the worker_id in the config
below by putting the name of the company running it so we can identify the exact work provided by your server.
config.toml:
Proving Keys generated by the ZKsync stack must be downloaded from https://pub-068f4ca2f03e4f21b6eb72b26008e90a.r2.dev/zksync-dev/data.tar.gz
and extracted on the host.
The docker image tag will be latest-89
for L4 and latest-86
for A10.