DeepSeek-v4-Flash 0731 DSpark Recipe

All projects

Apps & Tools

DeepSeek-v4-Flash 0731 DSpark Recipe

A high-performance recipe for serving DeepSeek-V4-Flash with 1M-token context on DGX Sparks.

byMia'a AI Lab

Open source

DeepSeek-v4-Flash 0731 DSpark Recipe media is blocked

Allow external media to connect to the provider and play this content.

Description

This repository provides a two-node DGX Spark recipe for serving DeepSeek-V4-Flash-Vision-Exp using vLLM with TP=2 and DSpark speculative decoding. It features native image support, a 1-million token context ceiling, and various performance hotfixes for long-context inference on GB10 clusters.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: A highly technical deployment setup proving extreme long-context (1M token) inference capabilities on specialized…
Attributed to: Mia'a AI Lab@MiaAI-Lab
Built with: V4.1 Flash Gemini 3.8 Flash Cyber 3.8 Flash Omni 1.1 Flash Ling 3.0 Flash Fin Gemini 3.7 Flash DeepSeek-V4-Flash-0731 Qwen3.7 Flash 3.5 Flash Cyber 3.6 Flash Step 3.7 Flash Gemini 3.5 Flash V4 2.5 Flash
Links: View tweet
Tags: #vllm #dspark #long-context #dgx-spark

Discovered: X
First posted: X · Aug 4, 2026
Listed: Added 41d ago
Build: Strong
Models: Strong