---
type: "Article"
title: "Qwen3.8-27B NVFP4 with TandemLLM on 1× DGX Spark"
description: "Qwen3.8-27B NVFP4 with TandemLLM on 1× DGX Spark: recipe by 0xBakeer with speed, condition and source for every number, weights and license."
resource: "https://www.contextstudios.ai/local-ai/0xbakeer--tandemllm-qwen38-27b-nvfp4-staircut-spark"
language: "en"
generated:
  by: "process:contextstudios-md/1"
  at: "2026-10-24T16:22:05.054Z"
status: "stable"
---

# Qwen3.8-27B NVFP4 with TandemLLM on 1× DGX Spark

Qwen3.8-27B NVFP4 with TandemLLM on 1× DGX Spark: 49.9 tok/s according to github.com (dataset as of Sep 29, 2026).

- Creator: [0xBakeer](https://www.contextstudios.ai/local-ai/creators/0xbakeer.md)
- Engine: TandemLLM v0.3.0 (eigene PyTorch/Triton/CUDA-Engine, StairCut)
- Quantization: NVFP4 (GPTQ) alle Projektionen + FP8-Head (über Qwen FP8-Checkpoint)
- Model family: Qwen3.8-27B
- Hardware: dgx-spark x1
- Context: 262144
- Artificial Analysis Intelligence Index: 33.7 ([source](https://artificialanalysis.ai/models/qwen3-8-27b))

## All numbers

- 49.9 tok/s (Mixed; Mixed, 1 request, benchmark prompt set, with DFlash2 drafter + StairCut (per recipe)) — [source](https://github.com/0xBakeer/TandemLLM#paper)
- decode_tps: 49.89 (1 request) — [source](https://github.com/0xBakeer/TandemLLM#paper)
- decode_tps: 90.45 (1 request) — [source](https://github.com/0xBakeer/TandemLLM#paper)

## Weights

- [0xBakeer/TandemLLM-Qwen3.8-27B-NVFP4](https://huggingface.co/0xBakeer/TandemLLM-Qwen3.8-27B-NVFP4) (main, apache-2.0)
- [Qwen/Qwen3.8-27B-FP8](https://huggingface.co/Qwen/Qwen3.8-27B-FP8) (base, apache-2.0)
- [0xBakeer/TandemLLM-Qwen3.8-27B-DFlash2-b16](https://huggingface.co/0xBakeer/TandemLLM-Qwen3.8-27B-DFlash2-b16) (draft, apache-2.0)
- [0xBakeer/TandemLLM-Qwen3.8-27B-DFlash2-b8](https://huggingface.co/0xBakeer/TandemLLM-Qwen3.8-27B-DFlash2-b8) (draft, apache-2.0)

[https://github.com/0xBakeer/TandemLLM](https://github.com/0xBakeer/TandemLLM)

## Related

- [Local AI on your hardware.](https://www.contextstudios.ai/local-ai.md)