​
Overview

These GPUs are validated for online RL and finetuning workflows. Availability may depend on your cloud provider.

​
Validated Types

  • H100 (80GB)
  • A100 (40GB/80GB)
  • A10G
  • L4

​
Notes

  • RDMA configurations are supported where available. See RL → Topology for GPU partitioning between vLLM, trainer, and reference models.
  • For best performance, match topology GPU counts with vLLM tensor parallel settings.