LMSYS Blog
LMSYS and Chatbot Arena research updates.
Read from its RSS feedAppears in:IA
Latest from LMSYS Blog
- SGLang security update: improving request isolationil y a 12 j
- More Than DeepInfer: What SGLang Deliversil y a 15 j
- SGLang Low-Latency Mode: Kernels and Runtimeil y a 21 j
- SGLang and Miles on NVIDIA Vera Rubinil y a 29 j
- Blog2026-09-24
- MiniMax-H3 on 8×H200: 1.95× Lossless, Up to 6.24× at 0.76–0.91 SSIM2026-08-27
- Qwen3.8-Flash-Next: Day-0 Support in SGLang2026-08-26
- Chasing the Batch-1 Floor: Ling-3.0-flash Speculative Decode on Blackwell2026-08-21
- Fast Engine Recovery: Sub-Second Engine Restart for SGLang via Weight Cache Daemon2026-08-21
- Mooncake for Miles: From Fragmented Rollout Data to Efficient Bulk I/O2026-08-20
- Pushing the Limits of Serving DeepSeek-V4-Pro2026-08-19
- Miles v0.1: Production-level Post-training2026-08-18
- Advanced CUDA Graph Techniques in SGLang2026-08-17
- Unified Radix Cache: One Tree for Hybrid Model Prefix Caching2026-08-11
- HPC-Ops × SGLang: High-Performance Attention, Router GEMM, and MoE Kernels from Tencent Hunyuan2026-08-07
- Towards Blackwell-Native 8-bit and 4-bit RL: End-to-End MXFP8 and NVFP4 RL in Miles2026-07-29
- SGLang and Miles Add Day-0 Support for Kimi K32026-07-27
- Serving GLM5.2 NVFP4 Agentic Workload with SGLang: Reaching 500 TPS in 2 Weeks2026-07-14
- DSpark in SGLang: Speculative Decoding with Confidence-Driven, Variable-Length Verification2026-07-06
- Agent-Assisted SGLang Development: An Initial Exploration2026-07-02
- The next generation of speculative decoding: DFlash and Spec V22026-06-15
- No Token Left Behind: Demystifying Token-In-Token-Out in Miles2026-06-05
- Higgs Audio v3 TTS on SGLang-Omni: Real-Time, Controllable Speech for Voice Agents2026-06-04
- DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles2026-04-25
- HiSparse: Turbocharging Sparse Attention with Hierarchical Memory2026-04-10
- Unlocking 25x Inference Performance with SGLang on NVIDIA GB300 NVL722026-02-20
- Deploying DeepSeek on GB300 NVL72: Big Wins in Long-Context Inference2026-02-19
- Pipeline Parallelism in SGLang: Scaling to Million-Token Contexts and Beyond2026-01-15
- EPD Disaggregation: Elastic Encoder Scaling for Vision-Language Models in SGLang2026-01-12
- Unified FP8: Moving Beyond Mixed Precision for Stable and Accelerated MoE RL2025-11-25