---
type: "Article"
title: "DeepSeek-V4-Flash 0731 MXFP4 with DwarfStar + DSpark on 2× DGX Spark (network TP)"
description: "DeepSeek-V4-Flash 0731 MXFP4 with DwarfStar + DSpark on 2× DGX Spark (network TP): recipe by antirez with speed, condition and source for every number, weights and license."
resource: "https://www.contextstudios.ai/local-ai/antirez--ds4-v4-flash-0731-mxfp4-dspark-tp-2x-spark"
language: "en"
generated:
  by: "process:contextstudios-md/1"
  at: "2026-10-24T16:15:04.870Z"
status: "stable"
---

# DeepSeek-V4-Flash 0731 MXFP4 with DwarfStar + DSpark on 2× DGX Spark (network TP)

DeepSeek-V4-Flash 0731 MXFP4 with DwarfStar + DSpark on 2× DGX Spark (network TP): 75.6 tok/s according to github.com (dataset as of Sep 29, 2026).

- Creator: [antirez](https://www.contextstudios.ai/local-ai/creators/antirez.md)
- Engine: DwarfStar (ds4) main@fc80bd69 (2026-10-07)
- Quantization: MXFP4 Experten (nativ), F16 HC/Compressor/Indexer, Q8 Attn/Shared/Out; Experten hälftig je Spark (~77 GiB je Host)
- Model family: DeepSeek-V4-Flash
- Hardware: dgx-spark x2
- Context: 65536
- Artificial Analysis Intelligence Index: 34.3 ([source](https://artificialanalysis.ai/models/deepseek-v4-flash))

## All numbers

- 75.6 tok/s (Peak; Peak, 1 request, code prompt, DSpark speculative decoding, 64K allocated) — [source](https://github.com/antirez/ds4/blob/main/docs/DGX_SPARK.md#flash-0731-on-two-sparks)
- decode_tps: 75.6 (1 request) — [source](https://github.com/antirez/ds4/blob/main/docs/DGX_SPARK.md#flash-0731-on-two-sparks)

## Weights

- [antirez/deepseek-v4-gguf](https://huggingface.co/antirez/deepseek-v4-gguf) (main, mit)
- [antirez/deepseek-v4-gguf](https://huggingface.co/antirez/deepseek-v4-gguf) (draft, mit)

[https://github.com/antirez/ds4](https://github.com/antirez/ds4)

## Related

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