# vllm-project/vllm-metal

Community maintained hardware plugin for vLLM on Apple Silicon

Repository: https://github.com/vllm-project/vllm-metal
Canonical: https://ross.abutalabs.com/products/vllm-metal
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-26T13:41:51+00:00

## Health v2 (maintenance only)
Score: 79/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 87, longevity 18
- inputs: {"age_days": 264, "days_push": 7, "days_rel": 7, "gap_med": 0, "n_releases_24m": 458}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1647, forks 228 (observed 2026-08-28T04:05:16.216288+00:00)

## What it is
A community-maintained hardware plugin that enables vLLM to run LLM inference on Apple Silicon Macs using MLX as the primary compute backend. It unifies MLX and PyTorch under a single lowering path with optimized Metal kernels for attention and prefill.

## Use cases
- serve LLMs locally on a Mac with Apple Silicon
- run vLLM inference on M-series chips
- host a 27B model on a single Mac
- use MLX as a backend for vLLM
- self-host an OpenAI-compatible LLM server on macOS
- benchmark LLM throughput on Apple hardware

## When to choose
- you want vLLM's serving stack on Apple Silicon instead of NVIDIA GPUs
- you need local LLM inference on macOS with high throughput
- you want to leverage MLX kernels and unified memory for large models

## When to avoid
- you are deploying on Linux servers with NVIDIA/AMD GPUs (use core vLLM)
- you need x86_64 or Rosetta support
- you run macOS earlier than 15 Sequoia or non-arm64 Python

## Facets
- artifact type: plugin
- maturity: active
- function: llm-inference, machine-learning, sdk
- domain: large-language-models, machine-learning, apple-ecosystem
- platform: python
- tags: vllm, mlx, apple-silicon, metal, hardware-plugin, llm-serving, macos, gpu

## Member repositories
- vllm-project/vllm-metal (main) score 79

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:16.216288+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:45:21.416473+00:00, confidence not recorded.
  - readme: https://github.com/vllm-project/vllm-metal (fetched 2026-08-28T04:05:16.216288+00:00, sha 0744ab80e626)
  - registry_pypi: https://pypi.org/pypi/vllm-metal/json (fetched 2026-08-29T11:18:44.650245+00:00, sha c2957ceb10a8)
- Data as of 2026-08-30T08:39:29.467469+00:00.