dgx-spark-qwen38
GitHub

dgx-spark-qwen38 / Models / Qwen3.8-Flash-Next 176B

Run Qwen3.8-Flash-Next 176B on a DGX Spark

A 176B mixture of experts on one box, by Qwen. NVFP4 with NEXTN speculation, on SGLang.

Qwen's 176B mixture of experts in NVFP4 on one box, at its full 262K window: its 47.7 GiB FP8 N-gram table lives in a sparse file on the local NVMe, so the memory goes to the weights and a real KV cache. On an ASUS Ascent GX10 with 128 GB: 47.9 tok/s, on code, with FLASH_TIER=context.

Measured

47.9tok/s

on code, with FLASH_TIER=context

  • A 262K-token window
  • GSM8K 97.41% on 1,314 problems
  • HumanEval 95.73%, MMLU 90.8%
Measured on
an ASUS Ascent GX10 with 128 GB, the project's reference box
How
the flash lane campaign
Source
docs/flash-lane.md

Load it

From the cockpit: open the Lanes view, pick Qwen3.8-Flash-Next 176B and press Load. From a terminal, in the repo folder on the box:

./switch-model.sh flash   # switches the boot, then prints the commands to run

switch-model.sh never restarts an engine itself: it prints the exact stop and start commands. The cockpit's Load runs them for you.

Other checkpoints of the same lane load the same way: flash-nvda, flash-uncensored.

On a 64 GB DGX Spark

No: it needs a 128 GB DGX Spark. A 64 GB box serves Nemotron 3.5 Bolt and Qwen3.8-27B: what runs on 64 GB.

Use it from your tools

The box serves OpenAI- and Anthropic-compatible APIs on port 30001, with an API key. Claude Code, opencode, Open WebUI and the OpenAI and Anthropic SDKs point there instead of a cloud: see Claude Code with a local model and opencode with a local model.

Install it

One command, on stock DGX OS. The same command updates it later.

curl -fsSL https://raw.githubusercontent.com/hasso5703/dgx-spark-qwen38/main/get.sh | bash

Questions about Qwen3.8-Flash-Next 176B on a DGX Spark

Can a DGX Spark run Qwen3.8-Flash-Next 176B?

Yes. Qwen's 176B mixture of experts in NVFP4 on one box, at its full 262K window: its 47.7 GiB FP8 N-gram table lives in a sparse file on the local NVMe, so the memory goes to the weights and a real KV cache. The installer of dgx-spark-qwen38 sets it up, and the cockpit loads it with one button.

How fast is Qwen3.8-Flash-Next 176B on a DGX Spark?

47.9 tok/s, on code, with FLASH_TIER=context, measured on an ASUS Ascent GX10 with 128 GB. A 262K-token window. GSM8K 97.41% on 1,314 problems. HumanEval 95.73%, MMLU 90.8%.

Does Qwen3.8-Flash-Next 176B run on a 64 GB DGX Spark?

No: it needs a 128 GB DGX Spark.

The other lanes