FR
Réserver un appel

Local AI · Recipe · 4× DGX Spark

DeepSeek-V4-Flash-0731 304.6B FP8 with vLLM on 4× DGX Spark

DeepSeek-V4-Flash-0731 304.6B FP8 with vLLM on 4× DGX Spark: 68.8 tok/s according to github.com (dataset as of Sep 29, 2026).

by fujitsupolycom

68.8tok/sEveryday

1 request, realistic prompt, with DSpark (per recipe)

Source: github.com
Intelligence (original model) · Artificial Analysis34with thinking
Instructions in the README Repo

Engine

Engine
vLLM
Quantization
Stock-Checkpoint (FP8 block-quantisiert, fp8_ds_mla KV)
Model family
DeepSeek-V4-Flash-0731
Context
1,048,576
Parameters
304.6B MoE (Summe der Safetensors-Tensoren lt. HF-API; 296.4B INT8-Experten)
Creator
fujitsupolycom
GitHub stars
82
Repo updated
Sep 27, 2026

Measurements

Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.

Decode · 1 request

  • Everyday (decode)68.8 tok/scontext 16K decode · Mittel aus 5 Beobachtungen (±20.05); unique cold prompts, temp 1.0; Workload-Klasse in der Quelle nicht klassifiziert Source 
  • Peak (decode)106 tok/scontext 2K decode · Mittel aus 5 Beobachtungen (±25.67) Source 

Prefill by context

  • 16K2,488 tok/sMittel aus 3 Beobachtungen, TTFT 6.586 s Source 

What you need

Hardware
4 × NVIDIA DGX Spark (GB10)
Engine
vLLM
Context
1,048,576 tokens

Notes

What matters before you rebuild it.

  • Custom kernel required

Sources

Related recipes

← Back to overview

Local AI in your company?

In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.

See the AI workshopBack to overview