---

license: mit

base_model: deepseek-ai/DeepSeek-V4.1-Flash

tags:

- gguf

- speculative-decoding

- dspark

- deepseek

- lucebox

---

DeepSeek V4.1 Flash DSpark drafter (GGUF)

The speculative-decoding drafter for DeepSeek V4.1 Flash in the

Lucebox engine. It is the draft block

DeepSeek ships inside the checkpoint (the mtp.* tensors: three stages, 128

routed experts top-3, a block of five tokens), converted to GGUF. The target

model drafts nothing itself; this file proposes tokens and the target verifies

them in one pass, and every verified token is bit-identical to plain decoding.

| file | size | sha256 |

|---|---:|---|

| DeepSeek-V4.1-Flash-DSpark-draft-MXFP4-Q8.gguf | 7.4 GiB | 3c96c3ee7801e5f166428b946e7590dbcf259b2ffe1fb09dd5fbed7d96bfe894 |

| DeepSeek-V4.1-Flash-DSpark-draft-Q2K-Q4K.gguf | 4.6 GiB | 5efffdb5ab4769bac7192fd2ce7c08cfa37945cad37be0825526c83139ef200a |

Routed experts are repacked bit-exactly to MXFP4, dense projections are Q8_0.

The Q2K-Q4K build requantizes the same drafter (routed experts Q2_K, dense

projections Q4_K) for the 192 GB setup, where the drafter shares the R9700 with

12 GB of hot experts: it leaves room for larger prompt chunks and drafts faster,

with about the same acceptance.

Converted with server/scripts/convert_ds4_dspark_draft_to_gguf.py from the

source checkpoint.

Run

With the target from

Lucebox/DeepSeek-V4.1-Flash-ROCMFP23-GGUF

on a Lucebox with 128 GB (AMD Radeon AI PRO R9700 + Ryzen AI MAX+ 395):

luce_server DeepSeek-V4.1-Flash-ROCMFP2S.gguf \
  --profile ds41-lucebox \
  --draft DeepSeek-V4.1-Flash-DSpark-draft-MXFP4-Q8.gguf

With 192 GB (Ryzen AI MAX+ PRO 495) every expert stays in memory, with the

smaller drafter and an expert placement and router bias fitted to the box:

luce_server DeepSeek-V4.1-Flash-ROCMFP2S.gguf \
  --profile ds41-gorgon \
  --draft DeepSeek-V4.1-Flash-DSpark-draft-Q2K-Q4K.gguf \
  --ds4-expert-placement placement.json \
  --ds4-router-bias router_bias.bin

Lucebox Manage installs the files and the fitted layout, and picks the launch

for the box's memory from its Engine page.

Measured

--profile ds41-gorgon with the Q2K-Q4K drafter, 192 GB, fresh server:

| prompt | prompt reading | decode |

|---|---|---|

| code, 36 tokens, 256 out | - | 49.4-49.5 tok/s |

| 2,027 tokens, 256 out | 811 tok/s | 32.0-32.6 tok/s |

| 30,752 tokens, 256 out | 1,039 tok/s | 31.8 tok/s |

--profile ds41-lucebox at its 128K context, first request after start

(DS41 guide):

| prompt | decode |

|---|---|

| code, 29 tokens, 256 out | 25.4-25.9 tok/s |

| 3,981 tokens, 128 out | 16.8-17.6 tok/s |

| 21,420 tokens, 32 out | 13.7 tok/s |

The engine verifies three drafted tokens per step by default (LUCE_DS4_SPEC_Q,

2 to 6), the fastest on the Lucebox.

License

MIT, as the source checkpoint.