#инференс
-
DeepSeek open-sources its inference engine DeepInfer
The company has released DeepInfer, the framework behind its production services, under Apache 2.0. It targets NVIDIA GPUs and Huawei Ascend NPUs.
-
Intel Unveils GPU D Series, Formerly Known as Crescent Island
Intel has announced its GPU D Series — enterprise-focused accelerators optimized for inference workloads. The products were previously known under the Crescent Island codename.
-
Can an AI Take Over the Server Through Inference-Engine Vulnerabilities?
An essay by Boyd Kane on how a malicious language model could gain control of the very machine hosting its own weights by exploiting bugs in engines like vLLM and SGLang.
-
Why a Local LLM Seems Dumber Than It Really Is
A poster on the Level1Techs forum ran a series of experiments showing that the very same model weights produce different tokens depending on the hardware and software — and that there's some very concrete math behind it.
-
AMD Buys Startup Taalas to “Etch” AI Models into Silicon
AMD has acquired Taalas — a company promising to speed up inference by baking neural network models directly into the fabric of a chip.
-
Kimi K3 on AMD MI355X: Better Cost per Token than NVIDIA B300
Wafer's engineers ran the open 2.8-trillion-parameter Kimi K3 model on AMD MI355X accelerators and got noticeably better performance per dollar than on NVIDIA's flagship Blackwell.
-
Baidu Releases vLLM-Kunlun — a Plugin for Running LLMs on Kunlun XPU Accelerators
Baidu has open-sourced vLLM-Kunlun, a pluggable module that lets you run the popular vLLM inference engine on its own Kunlun XPU accelerators without patching vLLM's source code.
-
Xiaomi Details Full-Pipeline Inference Optimization for the MiMo-V2.5 Series
Xiaomi has published a post on the MiMo project blog about full-pipeline inference optimization for the MiMo-V2.5 series of models. The details are laid out in the original publication.
-
GLM 5.2 on AMD MI355X: half the price of Blackwell at 2,626 tok/s per node
Wafer has announced the launch of the GLM 5.2 model on AMD Instinct MI355X accelerators, which delivered a performance of 2,626 tokens per second per node while more than halving the cost compared with solutions based on Nvidia Blackwell. The news drew a notable response on Hacker News.