Inference engine

SGLang

Serve supported language and multimodal models with prefix caching and multi-GPU inference.

Original Aguasto conceptual illustration. Not manufacturer geometry, an operating Aguasto asset, or a measured benchmark.

When it is relevant

  • Evaluate a serving engine for repeated-prefix or tool-use workloads
  • Study distributed inference with an exact checkpoint and parser
SGLang official documentation overview (external source) · SGLang hardware platforms (external source)

Hardware and software assumptions

Documented ecosystem: NVIDIA, AMD, Intel, Google, Huawei, Moore Threads. Platform support is not a guarantee that every checkpoint, optimization or feature works on every device.

  • Confirm model, tool-parser, precision and hardware support in the pinned release
  • Measure the actual prompt mix, concurrency, context and output lengths
  • Aguasto has not tested this stack or its performance
SGLang official documentation overview (external source) · SGLang hardware platforms (external source)

Operating trade-offs

  • Prefix-cache gains depend on reusable prompts and scheduling
  • A supported hardware platform does not imply every model and optimization works on it
SGLang official documentation overview (external source) · SGLang hardware platforms (external source)

Sources & verification

Reference revision 1 · Reviewed . Manufacturer claims and documentation are attributed; Aguasto has not run the cited products in production.

  1. SGLang official documentation overview ↗

    language and multimodal serving; RadixAttention and prefix caching; multi-GPU inference; documented hardware platforms. Publication: date not stated in the reviewed source. Verified: 2026-10-12.

  2. SGLang hardware platforms ↗

    Separate installation paths by hardware platform. Publication: date not stated in the reviewed source. Verified: 2026-10-12.