vLLM V1 backend parity restored after fixes
huggingface engineers restored backend parity between vllm v0.8.5 and v1.0.18.1 for reinforcement learning workloads. four specific fixes resolved discrepancies in rollout logprobs, runtime defaults, and weight update handling during grpo training.
Engineers achieved parity between vLLM V0.8.5 and V0.18.1 by resolving four critical backend discrepancies in an online RL setup. The fixes addressed processed rollout logprob semantics, explicit runtime defaults for prefix caching and async scheduling, and the inflight weight-update path. This established a stable baseline for V1, enabling objective-level corrections without semantic interference.