v1.0.0-rc0
Pre-release
Pre-release
·
42 commits
to main
since this release
Release candidate for first stable build of Arcee-NeMo-RL.
Features:
- Added first-class support for verifiers environments.
- Added a new DTensor "v2" backend for training with DTensor and 6D parallelism.
- Added a new vLLM-over-HTTP backend to support verifiers rollouts.
- Added native tool calling support to GRPO.