---
type: "Article"
title: "DeepSeek-V4.1-Flash-1017 EXL3 2.5 bpw with TensorFold on 2× DGX Spark"
description: "DeepSeek-V4.1-Flash-1017 EXL3 2.5 bpw with TensorFold on 2× DGX Spark: recipe by bertholomus with speed, condition and source for every number, weights and license."
resource: "https://www.contextstudios.ai/local-ai/bertholomus--deepseek-v4.1-flash-1017-tensorfold-exl3-2.5bpw-2x-dgx-spark"
language: "en"
generated:
  by: "process:contextstudios-md/1"
  at: "2026-11-11T03:30:11.033Z"
status: "stable"
---

# DeepSeek-V4.1-Flash-1017 EXL3 2.5 bpw with TensorFold on 2× DGX Spark

DeepSeek-V4.1-Flash-1017 EXL3 2.5 bpw with TensorFold on 2× DGX Spark: 60.5 tok/s according to github.com (dataset as of Nov 3, 2026).

- Creator: [bertholomus](https://www.contextstudios.ai/local-ai/creators/bertholomus.md)
- Engine: TensorFold native (Zig, TensorFold 1.0.12), BertholomusAI dsv41 family (bertholomus/TensorFold deepseek-v41-zig-tp2, recipe v0.7.3, served engine 6579b756dcd7) + kernel kit v0.7.0, TP=2 über RoCE, DSpark drafts
- Quantization: EXL3 2.5 bpw routed experts (2,52 bpw über die quantisierten Gewichte, ~185 GB; Attention/Shared Experts 4 bpw, DSpark-Drafter 4/6 bpw, Vision-Tower 16 bit)
- Model family: DeepSeek-V4.1-Flash-1017
- Hardware: dgx-spark x2
- Context: 1048576

## All numbers

- 60.5 tok/s (Everyday; Everyday, 1 request, prose prompt (greedy, thinking off), with DSpark (per recipe)) — [source](https://github.com/bertholomus/deepseek-v4.1-tensorfold-tp2-2xgb10#v073-2026-10-24-eight-concurrent-streams-the-same-replies)
- decode_tps: 60.5 (1 request) — [source](https://github.com/bertholomus/deepseek-v4.1-tensorfold-tp2-2xgb10#v073-2026-10-24-eight-concurrent-streams-the-same-replies)
- decode_tps: 159.5 (1 request) — [source](https://github.com/bertholomus/deepseek-v4.1-tensorfold-tp2-2xgb10#v073-2026-10-24-eight-concurrent-streams-the-same-replies)
- prefill_tps: 2364 (1 request) — [source](https://github.com/bertholomus/deepseek-v4.1-tensorfold-tp2-2xgb10#v073-2026-10-24-eight-concurrent-streams-the-same-replies)

## Weights

- [bertholomus/DeepSeek-V4.1-Flash-1017-EXL3-2.5bpw](https://huggingface.co/bertholomus/DeepSeek-V4.1-Flash-1017-EXL3-2.5bpw) (main, mit)
- [deepseek-ai/DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017) (base, mit)

[https://github.com/bertholomus/deepseek-v4.1-tensorfold-tp2-2xgb10](https://github.com/bertholomus/deepseek-v4.1-tensorfold-tp2-2xgb10)

## Related

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