---
type: "Article"
title: "DeepSeek-V4.1-Flash-1017 EXL3 3.5 bpw with TensorFold on 4× DGX Spark"
description: "DeepSeek-V4.1-Flash-1017 EXL3 3.5 bpw with TensorFold on 4× DGX Spark: recipe by tonyd2wild with speed, condition and source for every number, weights and license."
resource: "https://www.contextstudios.ai/local-ai/tonyd2wild--deepseek-v4.1-flash-1017-tensorfold-exl3-3.5bpw-4x-dgx-spark"
language: "en"
generated:
  by: "process:contextstudios-md/1"
  at: "2026-11-11T03:30:16.027Z"
status: "stable"
---

# DeepSeek-V4.1-Flash-1017 EXL3 3.5 bpw with TensorFold on 4× DGX Spark

DeepSeek-V4.1-Flash-1017 EXL3 3.5 bpw with TensorFold on 4× DGX Spark: 78 tok/s according to github.com (dataset as of Nov 3, 2026).

- Creator: [tonyd2wild](https://www.contextstudios.ai/local-ai/creators/tonyd2wild.md)
- Engine: TensorFold (bertholomus deepseek-v41-zig-tp4 @ 5940208, recipe kit v0.1.1) + 4 patches, RoCE rail 1 (config.env)
- Quantization: EXL3 3.5 bpw, all matrices (LibertAI mul1)
- Model family: DeepSeek-V4.1-Flash-1017
- Hardware: dgx-spark x4
- Context: 1048576

## All numbers

- 78 tok/s (Everyday; Everyday, 1 request, prose prompt, with DSpark (per recipe)) — [source](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks/blob/main/runs/2026-10-17-1017-tensorfold/results/tf1017a/bench-tf1017a.md)
- decode_tps: 78.04 (1 request) — [source](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks/blob/main/runs/2026-10-17-1017-tensorfold/results/tf1017a/bench-tf1017a.md)
- decode_tps: 89.78 (1 request) — [source](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks/blob/main/runs/2026-10-17-1017-tensorfold/results/tf1017a/bench-tf1017a.md)
- decode_tps: 134.76 (1 request) — [source](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks/blob/main/runs/2026-10-17-1017-tensorfold/results/tf1017a/bench-tf1017a.md)
- prefill_tps: 1928.6 (1 request) — [source](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks/blob/main/runs/2026-10-17-1017-tensorfold/results/tf1017a/bench-tf1017a.md)
- ttft_ms: 191 (1 request) — [source](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks/blob/main/runs/2026-10-17-1017-tensorfold/results/tf1017a/bench-tf1017a.md)

## Weights

- [LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-3.5bpw](https://huggingface.co/LibertAIDAI/DeepSeek-V4.1-Flash-1017-EXL3-3.5bpw) (main, mit)
- [deepseek-ai/DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) (base, mit)

[https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks](https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks)

## Related

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