# 推理专题日跟踪 · 2026-09-11 > 日期:2026-09-11(Asia/Shanghai)。专题正文:`intel/topics/inference.md`。原料:`archive/2026-09-11/`。 ## 摘要 - **新稿 · Nutanix Private Inference:** 张量并行多 GPU、批推理、投机解码宣称至 2.5×、LoRA <8B(**厂商自称**);与 NAI Agent Gateway 同发。 - **续档 · RH AI 3.5:** priority-aware serving / vLLM CPU·NVMe KV offload / llm-d 多云 GA·TP。 - **续档 · Rafay×RH Token Factory:** 多租户计量推理云参考架构。 ## 原文摘录 ### Nutanix Private Inference - 归档:`industry/nutanix-agentic-ai-gec-2026-09-09.txt` > … scalable, multiGPU inference for LLMs via tensor parallelism… Speculative decoding which accelerates LLM inference token generation by up to 2.5x… LoRA fine-tuning for smaller models (<8B parameters)…(**厂商自称**) ### RH AI 3.5 · 续档 - `archive/2026-09-11/redhat/red-hat-ai-3-5-scaling-and-governing-ai-agents-in-production-0f281ed9.txt` > … priority-aware serving (GA)… vLLM introduces CPU offloading (GA) and NVMe storage offload for the KV cache… ## 检索范围 - archive/2026-09-11/industry(Nutanix)、redhat;对照 2026-09-10 inference 日跟踪