OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

Hummingbird — prayangshuuu

Hummingbird is a C17 runtime for local inference of open-source dense and Mixture of Experts language models. It treats SSD, RAM and VRAM as a managed memory hierarchy so models larger than available accelerator memory can be used on consumer hardware.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Hummingbird — prayangshuuu — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Hummingbird when you want to build a local inference runtime and are interested in placing model weights across SSD, RAM and VRAM rather than requiring all weights to fit in accelerator memory.

Who it is for

  • Developers building or evaluating a self-hosted language-model inference runtime.
  • Users seeking local inference for open-source models on hardware with limited RAM or accelerator memory.
  • Engineers who want a C17 runtime with a CPU reference backend and documented CLI frontends.

Documented uses

  • Generate text from a local model file with the `hb` CLI.
  • Use interactive chat mode with a local model.
  • Start an OpenAI-compatible HTTP server around a local model.
  • Build and test the runtime from source.
Before you choose

Requirements & limits

The README documents a Linux/macOS installer and a source build using CMake. The source build lists GCC, Clang or MSVC as compiler options. The CPU backend is enabled by default; CUDA and Metal are optional build settings marked as planned.

  • The memory hierarchy streams routed expert weights from disk while keeping frequently needed weights resident and caching recent weights.
  • The project documents a GPT-OSS adapter as shipped; other named families should be checked against the current repository before selecting a model.
  • The README lists CUDA and Metal acceleration as planned, with both build options disabled by default.
  • The documented CLI accepts model files with the .hbm suffix in its usage examples.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Build Hummingbird from source

Compile the runtime and its default frontends from the documented development preset.

Before you start

  • A C17 compiler such as GCC, Clang or MSVC
  • CMake 3.20 or newer
  • The Hummingbird repository

Steps

  1. Clone the repository and change into its directory.

  2. Configure the development preset with cmake --preset dev.

  3. Build with cmake --build --preset dev.

  4. Run the documented test suite with ctest --preset dev.

Expected outcome

A source build of Hummingbird; the README documents the `hb` CLI and other frontends among the build options.

Guide limitations

  • This guide covers building and the documented test command, not running inference.
  • CUDA and Metal backends are off by default and marked as planned in the README.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
prayangshuuu/hummingbird · Official homepage title · retained in its source language
How we review sourcesSuggest a correction