
The biggest win for our team is the developer experience. Instead of waiting around for cloud instances I can test my Ray code right on my laptop. Once it works locally, pushing it to a cluster of cloud nodes is essentially seamless. The built-in workspaces feel exactly like my usual IDE, which made onboarding incredibly fast. We use it heavily for LLM training. Review collected by and hosted on G2.com.
Two things stand out. First, tracking down out of memory errors across a massive cluster can be a headache. The logs are okay, but finding the exact point of failure takes some digging. Second, keeping an eye on cloud spend is tricky. The auto-scaling spins up resources so aggressively that if you aren't paying attention, your cloud budget can jump unexpectedly. Review collected by and hosted on G2.com.