AI Briefing#sglang
The latest AI and developer news about #sglang, with the original source and a short summary.
The latest AI and developer news about #sglang, with the original source and a short summary.
RadixArk has released Miles, an open-source framework that integrates SGLang, Megatron-LM, and Ray for large-scale LLM reinforcement learning (RL) post-training.
DFlash technology, developed jointly by Z Lab and Modal, achieved LLM inference performance far superior to existing MTP methods.
OSCAR, a technique using spectral covariance-aware rotation to reduce quantization loss in KV Cache, has been released.
An analysis alleges that Alibaba T-Head's asllm engine uses the open-source sglang without authorization and has left a serious bug unaddressed.
Qwen3.7-Max has unveiled its agentic, coding, and long-horizon autonomous execution performance.
Amazon extended the Chinchilla scaling law to present a framework that speeds up LLMs without losing accuracy.
MiniCPM has released MiniCPM-V 4.6, unveiling a 1.3B vision-language model for mobile use.
SGLang has been found to have FP8 KV cache corruption and a memory leak in image requests.
This summarizes the major technical issues and solutions that arise during fine-tuning and deployment of the Gemma-4 model.
In an apples-to-apples FP8 comparison, the advantage between Ironwood and B200 reversed depending on the model.
Qwen3.6 is the latest Qwen model family with strengthened agentic coding and thinking-context preservation.
A diffusion LLM has achieved both AR-level quality and parallel generation speed together.
I-DLM simultaneously boosts DLM quality and speed through introspective consistency.
The nondeterminism of LLM inference is less about GPU concurrency and more about request scheduling.
Verified Qwen3.5-122B running at 198 tok/s on 2x RTX PRO 6000 Blackwell.
SGLang now supports the Transformers backend, enabling high-performance inference for any Transformers-compatible model.