#sglang
Sglang: 14 AI articles covering sglang news, analysis, and research
Articles
Infini-AI-Lab Releases Hermit: An Open-Source Harness Adapterโญ9
Infini-AI-Lab releases Hermit, an open-source harness adapter framework that adapts AI agents like Codex and Claude Code to new tasks without fine-tuning models...
Polymathic AI and Princeton Release FlashMap: A Flow-Map Drafterโญ8
FlashMap, a flow-map speculative decoding drafter from Polymathic AI and Princeton, drafts 16-token blocks in one pass for up to 6.14x faster, lossless LLM deco...
Tsinghua AI Open-Sources AKV: A Cache Engineering Server for LLMโญ9
Tsinghua researchers open-source AKV, a KV cache engineering server on SGLang that cuts LLM agent inference costs by 27.7% while improving task scores by 1.7 po...
Nanbeige Releases Nanbeige4.3-33B: A 33B Open Agentic Coding Modelโญ8
Nanbeige4.3-33B is a 33B open agentic coding model with a 512K context window, Apache 2.0 license, and strong benchmark results against near-30B rivals.
Google Cloud Releases llm-d 1.0: Cache-Aware Routing Across TPUโญ8
Google Cloud releases llm-d 1.0, an open-source Kubernetes-native inference stack with cache-aware routing across TPU and GPU fleets for distributed AI serving.
Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learningโญ9
Alibaba's Qwen team introduces QwenGyre, an elastic reinforcement learning framework for extreme-long-horizon agents, delivering up to 1.85x speedup on long cod...
Can an Open Model Do Security Research? Cantina's apex-flash-1โญ7
Cantina's open-weights apex-flash-1, a fine-tuned GLM-5.3-Flash model, solves 40 of 60 held-out bug tasks for vulnerability research.
Liquid AI Releases LFM2.5-VL-3B-DSpark: Speculative Decoding forโญ8
Liquid AI releases LFM2.5-VL-3B-DSpark, a speculative decoding drafter for its vision-language model, delivering up to 3.13x faster decoding on Apple silicon, w...
BottleCap AI Releases ThinkingCap-Qwen3.8-27B: 37.2% Fewerโญ8
ThinkingCap-Qwen3.8-27B by BottleCap AI cuts thinking tokens by 37.2% on average across 12 benchmarks, with only a 0.86pp accuracy drop.
Alibaba Qwen Releases Qwen-Image-2.1: A 7B Open-Weight Model forโญ8
Alibaba Qwen releases Qwen-Image-2.1, a 7B open-weight model unifying text-to-image generation and editing in one checkpoint, with Day 0 support and mixed-granu...