# eugr/spark-vllm-docker

Docker configuration for running VLLM on dual DGX Sparks

Repository: https://github.com/eugr/spark-vllm-docker
Canonical: https://ross.abutalabs.com/products/spark-vllm-docker
Language: Shell
License: MIT
License Family: permissive
Last push: 2026-08-26T12:12:24+00:00

## Health v2 (maintenance only)
Score: 83/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 20
- inputs: {"age_days": 281, "days_push": 7, "days_rel": 7, "gap_med": 1, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2189, forks 370 (observed 2026-08-28T04:06:23.722555+00:00)

## What it is
A Docker configuration and set of shell scripts for running vLLM inference on NVIDIA DGX Spark hardware, from single nodes to multi-node clusters using Ray or PyTorch distributed. It includes support for InfiniBand/RDMA networking, fast model loading, and tested nightly Docker images.

## Use cases
- run vllm on dgx spark
- set up multi-node llm inference cluster
- serve large language models across two dgx sparks
- dockerize vllm with infiniband support
- deploy vllm with ray distributed backend
- benchmark llm inference on spark hardware
- download and load models fast with fastsafetensors

## When to choose
- you own one or more DGX Spark machines and want optimized vLLM inference
- you need multi-node LLM serving with InfiniBand/RDMA or QSFP networking
- you want pre-tested nightly Docker images instead of hand-building vLLM
- you need cluster orchestration via Ray or native PyTorch distributed mode

## When to avoid
- you run inference on non-DGX-Spark GPUs or cloud instances
- you need a general-purpose vLLM deployment on Kubernetes
- you want a managed inference service rather than self-hosted infrastructure
- you need training or fine-tuning rather than inference

## Facets
- artifact type: infra-config
- maturity: active
- function: llm-inference, container-runtime, deployment, gpu-computing, developer-tools
- domain: large-language-models, gpu-computing, infrastructure-as-code, self-hosted
- platform: self-hosted
- tags: vllm, dgx-spark, multi-node, infiniband, nccl, ray, docker-compose, nvidia, containers, docker, linux, gpu

## Member repositories
- eugr/spark-vllm-docker (main) score 83

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:23.722555+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 2026-08-30T02:47:43.752720+00:00, confidence not recorded.
  - readme: https://github.com/eugr/spark-vllm-docker (fetched 2026-08-28T04:06:23.722555+00:00, sha fc31885bb6a9)
- Data as of 2026-08-30T08:39:29.467469+00:00.