vLLM ships Day-0 LoRA RL for Muse Spark Open, lifting AIME from 0.56 to 0.85
vllm_project · x · 2026-10-16
The vLLM team announced Day-0 LoRA RL support for Muse Spark Open, using vLLM for rollouts and verl FSDP2 for training.
- Architecture: vLLM serves the NVFP4 base with LoRA; the trainer uses a BF16 dequantization of the same weights; R3 replays rollout routing, and only the adapter is updated and synced back
- Results: GRPO on DAPO-Math-17K, 150 steps on 16 GB300 GPUs, improved AIME-2024 mean@8 from 0.56 to 0.85, with train–rollout k3 KL near 2×10⁻⁴
- Titan RL also uses vLLM as its rollout engine for Day-0 RL
- Part of broader cross-hardware serving work including fused FlashInfer kernels and AMD MI355X FP8/MXFP4 support
More from Infra
- Temporal Serverless Workers on Cloud Run: queue-driven autoscaling, scale to zero — rseroter · 2026-10-16
- LMA brings day-0 model support to TPU for the first time, spanning NVIDIA, AMD, Intel via SGLang — ying11231 · 2026-10-16
- Semtech's analyst-day slide cleanly explains the four signal-integrity impairments — BenBajarin · 2026-10-16
- Fortytwo Engine boosts Qwen3.8-27B accuracy 7% and cuts tokens 25% via Swarm Inference — rohanpaul_ai · 2026-10-16
- Berkeley scholars: concentrated chip supply chain makes a hardwired frontier AI pause feasible — wfithian · 2026-10-16
- Robotics data tokenization landscape: Khala maps peaq, THOR, BitRobot and Axis — 0xSammy · 2026-10-16