# MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark

DeepSeek-v4-Flash 0731 recipe for 2x DGX Sparks

Repository: https://github.com/MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark
Canonical: https://ross.abutalabs.com/products/deepseek-v4-flash-dspark-2x-dgx-spark
Homepage: https://x.com/MiaAI_lab
Language: Python
License: MIT
License Family: permissive
Last push: 2026-09-02T19:54:34+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 35, longevity 4
- inputs: {"age_days": 66, "days_push": 0, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1176, forks 165 (observed 2026-09-03T02:15:06.654273+00:00)

## What it is
A deployment recipe for serving DeepSeek-V4-Flash-0731 across two NVIDIA DGX Spark nodes using vLLM with tensor parallelism and DSpark speculative decoding. It provides environment configuration, weight caching scripts, and validation tooling for a two-node inference cluster.

## Use cases
- serve deepseek v4 flash on two dgx spark nodes
- run vllm with tensor parallelism across multiple machines
- set up speculative decoding for faster llm inference
- deploy a large language model on a small gpu cluster
- configure nccl networking for multi-node inference
- self-host deepseek models with 1m token context

## When to choose
- you own two DGX Spark devices and want a tested serving recipe
- you need high-throughput DeepSeek V4 Flash inference with long context
- you want preconfigured vLLM multi-node deployment scripts

## When to avoid
- you have a single GPU or non-DGX hardware
- you need training or fine-tuning rather than inference
- you want a general-purpose serving stack rather than a model-specific recipe

## Facets
- artifact type: infra-config
- maturity: active
- function: llm-inference, deployment, gpu-computing, developer-tools
- domain: large-language-models, gpu-computing, self-hosted, infrastructure-as-code
- platform: self-hosted, python
- tags: vllm, dgx-spark, speculative-decoding, deepseek, multi-node, tensor-parallelism, recipe, linux, docker, gpu

## Member repositories
- MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:06.654273+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 2026-08-30T07:13:41.869514+00:00, confidence not recorded.
  - readme: https://github.com/MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark (fetched 2026-09-03T02:15:06.654273+00:00, sha dad0c7cd78e6)
- Data as of 2026-08-30T08:39:29.467469+00:00.