在 NVIDIA DGX Spark 上快速启动、管理和停止 LLM 推理服务，支持 vLLM、SGLang、llama.cpp 等后端，省去手动配置环境的麻烦。
在 DGX Spark 上跑 Qwen3.5-122B-A10B 的 INT4 量化推理方案，靠 AutoRound 加 MTP 把吞吐从 28.3 提到 51 tok/s，显存吃紧的单机也能跑大模型。
面向 Mac 和 NVIDIA DGX Spark 的 AI 电影制作工作室，按剧组流程规划分镜，并用 Qwen-Image、FLUX、MiniMax 等模型生成带声音的画面与视频。
把文档解析、知识图谱和向量检索串成一套本体增强的 RAG 方案，让 AI Agent 基于结构化知识作答，减少胡编乱造。
纯 Rust 实现的大模型推理引擎，支持 Transformer、Mamba/SSM、投机解码和 NVFP4 量化，兼容 OpenAI API，适合在 DGX 等硬件上部署高效推理服务。
基于 Qwen3.6-27B 做去审查与能力增强的 abliteration 模型，配合 NVFP4 量化和 DFlash 投机解码，在 DGX Spark 上优化长上下文草稿接受率，附多种 HF 变体与 Docker 部署。
一条命令把 NVIDIA DGX Spark 或任意 GB10 机器（华硕、戴尔、惠普等）变成私有 AI 服务器，内置实测模型通道、OpenAI/Anthropic 兼容 API 和本地控制台。
…-11-07 with the command of this release, cd ~/dgx-spark-qwen38 && curl -fsSL https://get.alzeo.ai bash , once GitHub's CI had passed on that commit (7 jobs, 89 steps, non
Two boxes behind one endpoint. Two DGX Spark that each run Alzeo can answer as one (discussion 45). One box can keep the 27B for the agents while the other serves Flash o
… systems, including Jetson, Grace Hopper, and DGX Spark. - 🧩 Simpler OpenVINO setup: Uses prebuilt ONNX Runtime packages with OpenVINO for Linux and Windows x64, replaci
…zed environments 'out of the box', including: DGX Spark, AMD GPU, Windows CUDA (4090+), Linux CUDA (8.9+), along with the standard binaries for Windows x86, Windows ARM64
…s to @odnodn for adding (experimental) Nvidia DGX Spark support in 292! Thanks to @andrewmc201 for his additions around Live Mode and CI in 299, 300 and 301! Thanks to @N
…ardware platforms and improves performance on RTX Spark / DGX Spark. Highlights Support added for NVIDIA RTX Spark products running Windows on ARM64 ( 2038, 2037, 2032) I