IT
Prenoti una chiamata

Local AI · Recipe · 2× DGX Spark

Qwen3.8-Flash-Next INT4-AutoRound (v5.2) with vLLM on 2× DGX Spark

Qwen3.8-Flash-Next INT4-AutoRound (v5.2) with vLLM on 2× DGX Spark: 74 tok/s according to github.com (dataset as of Sep 29, 2026).

by myllmbox

74tok/sEveryday

1 request, prose prompt (long story), thinking off, with MTP

Source: github.com
Intelligence (original model) · Artificial Analysis40with thinking

Install: Qwen3.8-Flash-Next INT4-AutoRound (v5.2) with vLLM on 2× DGX Spark

Command
git clone https://github.com/myllmbox/qwen38-flash-next-cluster-recipe.git && cd qwen38-flash-next-cluster-recipe && ./run.sh

Full instructions in the README Repo

Engine

Engine
vLLM 0.30 (Image myllmbox/qwen38-flash-next-cluster-vllm:v5.2; RecoverSSM, dynamischer Draft-Depth 3–7 in Full-CUDA-Graphs, sampled Drafts + Block-Verifikation, One-Shot-RoCE-All-Reduce, Dual-Rail, TP=2, 64 Seats) — Kit v5.2
Quantization
INT4-AutoRound (azampatti): int4-Experten, fp8-Seitenschichten; n-gram-Tabelle beim Laden fp8 → NVFP4, halb pro Box auf der GPU; KV-Pin 29G/Box (1.829.182 Tokens)
Model family
Qwen3.8-Flash-Next
Context
262,144
Parameters
125B-A5B (Modellname); 5 von 512 Experten pro Token
Creator
myllmbox
GitHub stars
127
Repo updated
Oct 6, 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)74 tok/sLong prose (7.000-Wörter-Story), c=1, Thinking off, Mittel aus 3 Runs Source 
  • Mixed (decode)100 tok/sMixed (code + explanation), c=1, Thinking off, Mittel aus 3 Runs; Code (TypeScript) 133,6, structured 124,4 Source 
  • Peak (decode)177 tok/sBestes 10-s-Fenster bei c=1 (Thinking-on-Request: 93,2 Ø / 176,8 peak); aggregiert c=64: 1.174,0 peak / 968,4 Ø (mixed) Source 

Prefill by context

Time to first token

Shorter is better.

  • Prompt not stated360 mscontext 1K · First token, 1k prompt: 0,36 s Source 

What you need

Hardware
2 × NVIDIA DGX Spark (GB10)
Engine
vLLM 0.30 (Image myllmbox/qwen38-flash-next-cluster-vllm:v5.2; RecoverSSM, dynamischer Draft-Depth 3–7 in Full-CUDA-Graphs, sampled Drafts + Block-Verifikation, One-Shot-RoCE-All-Reduce, Dual-Rail, TP=2, 64 Seats) — Kit v5.2
Context
262,144 tokens

Notes

What matters before you rebuild it.

  • Custom kernel required
  • License unclear

Sources

Related recipes

2× DGX Spark

Qwen3.8-Flash-Next 125B INT4-AutoRound with TensorFold on 2× DGX Spark (1M context)

by MiaAI-Lab

89.3tok/sEveryday

1 request, prose prompt, with MTP (per recipe) Source 

Intelligenceno independent value
  • Custom kernel required
  • License unclear

INT4-AutoRound (GPTQ int4, Top-5-Experten) + FP8 n-gram-Tabelle, FP8 KVTensorFold tensorfold-native (Zig, zig-flashnext db281878 + patches), Image ghcr.io/miaai-lab/qwen3.8-flash-dual-dgx-sparks-tensorfold:zig-db28187-f3e1300f4f62, TP=2 über CX7/RoCE, MTP + Copy-Drafts (QUANT=int4ar)Weights Repo Updated Oct 8, 2026

Details
2× DGX Spark

Qwen3.8-Flash-Next 125B NVFP4 with TensorFold on 2× DGX Spark (1M context)

by MiaAI-Lab

67.4tok/sEveryday

1 request, prose prompt, with MTP (per recipe) Source 

Intelligence40with thinking
  • Custom kernel required
  • License unclear

NVFP4 (nvidia, Experten) + FP8 n-gram/MTP, FP8 KVTensorFold tensorfold-native (Zig, zig-flashnext db281878 + patches), Image ghcr.io/miaai-lab/qwen3.8-flash-dual-dgx-sparks-tensorfold:zig-db28187-f3e1300f4f62, TP=2 über CX7/RoCE, MTP + Copy-DraftsWeights Repo Updated Oct 8, 2026

Details

← 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