vLLM achieves bitwise-consistent RL training, keeps 91% rollout throughput, open-sources agent skill

vLLM Blog · rss · 2026-10-14

The vLLM team made the rollout and training forward pass produce bitwise-identical logprobs, eliminating train-inference mismatch in RL post-training.

• Verified on DeepSeek-V4-Flash and Nemotron-3.5-Lightning-30B-A3B (NVFP4)
• Retains 91% of rollout throughput on DeepSeek-V4-Flash
• The accompanying agent skill is open-sourced for direct reuse

Original post →

More from Infra

• OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
• Jensen Huang lays out NVIDIA's full Vera Rubin agentic system: NVL72, MGX, CMX and RTX — rohanpaul_ai · 2026-10-23
• ONNX to Transformers.js: running models directly in your browser — JosephJacks_ · 2026-10-23
• MiniMax M3.1 Flash free for 3 days on 0G, with 1M context and 512K output — MiniMax_AI · 2026-10-23
• "We bet on Cloudflare and it paid off": R2 and D1 at billion scale — dinasaur_404 · 2026-10-23
• Public data is drying up; private training data is turning into a browsable catalog — SucceededMind · 2026-10-23

Infra channel →