面向大语言模型的高吞吐、低显存推理与服务引擎，通过 PagedAttention 等优化显著提升并发吞吐并降低显存占用，适合部署在线推理服务。
给 LLM 推理加速的 KV Cache 层，可对接 vLLM 等框架，复用和缓存键值对来降低显存占用、提升吞吐，支持 CUDA 和 ROCm。
高性能大模型与多模态模型推理服务框架，通过优化注意力机制和调度策略提升吞吐与延迟表现，支持 Llama、DeepSeek、MoE 等主流模型部署。
…e-darwin-arm64.dmg - macOS (Intel): magpie-darwin-amd64.dmg - Windows: magpie-windows-amd64.exe (ARM: magpie-windows-arm64.exe ) - Linux: magpie-linux-amd64 / magpie-linu
…r load 导入，配套部署 ZIP 包含 Compose 与部署/API说明。仅支持 linux/amd64；需要自行配置随机令牌与可写的已有挂载目录。TV不随本版发布，个人手机版保持独立。 Windows EXE 未签名。实体 Intel、实体 Windows/GPU、飞牛硬件与 Docker ARM64 不在本轮验收范围；不承诺上游
…mplete information, please refer to the official [AMD Security Bulletin](https://www.amd.com/en/resources/product-security/bulletin/amd-sb-6032.html). Behavioral Changes
…hanges - 🚀 New accelerator support: Adds working AMD MIGraphX provider registration, selectable with migraphx:0 , and prebuilt CUDA, TensorRT, and NVRTX runtime builds f
…time GPU utilization ( NVIDIA via nvidia-smi , or AMD via sysfs ) to the system monitor, displayed in the top bar alongside the CPU and in the monitoring menu. - On NVIDI
…See [ROCm compatibility matrix](https://rocm.docs.amd.com/en/docs-6.4.2/reference/gpu-arch-specs.html) for the full list of GPU architectures. Every architecture listed t