DeepSeek-V4.1 Flash MXFP4 GGUF for DwarfStar

GGUF files for DeepSeek V4.1 Flash in DwarfStar (DS4). They work with upstream main and with the kernelpool/ds4:ds41-perf branch of antirez/ds4#1073, which speeds up V4.1 decoding and prefill on Metal. These are DS4 files, not llama.cpp GGUFs: llama.cpp does not load DS4's deepseek41 layout.

The files come from the original September release; files for DeepSeek-V4.1-Flash-1017 are in kernelpool/DeepSeek-V4.1-Flash-1017-MXFP4-GGUF. Keep the main model, the drafter and the vision encoder from the same release: the 1017 checkpoint has different weights, including the DSpark stages and the vision tower.

File	Size	Content
`DeepSeek-V4.1-Flash-MXFP4.gguf.part1`, `.part2`	467.2 GiB	Main model, `deepseek41` layout, in two parts: routed experts are the checkpoint's own MXFP4 blocks repacked without changing their values; Q8_0 attention, shared experts and output head; F16 embeddings; the two Engram tables in the checkpoint's FP8 encoding (188.8 GiB, read from disk)
`DeepSeek-V4.1-Flash-DSpark.gguf`	8.0 GiB	DSpark drafter for speculative decoding, `deepseek41-dspark` layout: native MXFP4 experts, dense weights expanded losslessly to F16 or F32
`DeepSeek-V4.1-Flash-Vision.gguf`	0.9 GiB	Vision encoder for image input, `deepseek4-vision` layout: the checkpoint's vision tower and aligner in BF16

The drafter replaces DeepSeek-V4.1-Flash-DSpark-MXFP4.gguf, an older format that current DS4 does not load; that file remains at revision 22a073e5.

Run

Join the two parts first (the parts can be deleted afterwards):

cat DeepSeek-V4.1-Flash-MXFP4.gguf.part1 DeepSeek-V4.1-Flash-MXFP4.gguf.part2 > DeepSeek-V4.1-Flash-MXFP4.gguf

Build either tree, then run:

git clone https://github.com/antirez/ds4 && cd ds4 && make                    # upstream
git clone -b ds41-perf https://github.com/kernelpool/ds4 && cd ds4 && make    # antirez/ds4#1073

./ds4 -m DeepSeek-V4.1-Flash-MXFP4.gguf
./ds4 -m DeepSeek-V4.1-Flash-MXFP4.gguf --dspark --mtp-model DeepSeek-V4.1-Flash-DSpark.gguf
./ds4 -m DeepSeek-V4.1-Flash-MXFP4.gguf --vision DeepSeek-V4.1-Flash-Vision.gguf
./ds4-server -m DeepSeek-V4.1-Flash-MXFP4.gguf --dspark --mtp-model DeepSeek-V4.1-Flash-DSpark.gguf

The main model keeps 278 GiB of weights resident and reads the Engram tables from disk; the drafter adds 8 GiB. Tested on a 512 GB Mac Studio M3 Ultra. DSpark is optional and does not help every prompt; see speculative decoding. With --vision, /read FILE in ./ds4 submits a PNG or JPEG.

Conversion

Written from deepseek-ai/DeepSeek-V4.1-Flash at revision dba1be0a40aa45a94ad051997016db3960a90277, which the main model and drafter record in their metadata; gguf-tools/deepseek41_vision.py always records df42c109, a commit from the same day with identical weights. The MXFP4 mode of gguf-tools/deepseek41_quantize.py is in the ds41-perf branch (upstream's converter writes Q2 and Q4); gguf-tools/deepseek41_dspark.py and gguf-tools/deepseek41_vision.py are in both trees.

make -C gguf-tools libds4quants.dylib
python3 gguf-tools/deepseek41_quantize.py --hf models/DeepSeek-V4.1-Flash \
  --source-revision dba1be0a40aa45a94ad051997016db3960a90277 \
  --quant mxfp4 --out DeepSeek-V4.1-Flash-MXFP4.gguf
python3 gguf-tools/deepseek41_dspark.py --hf models/DeepSeek-V4.1-Flash \
  --source-revision dba1be0a40aa45a94ad051997016db3960a90277 \
  --out DeepSeek-V4.1-Flash-DSpark.gguf
python3 gguf-tools/deepseek41_vision.py --hf models/DeepSeek-V4.1-Flash \
  --out DeepSeek-V4.1-Flash-Vision.gguf

Quality

Scored with gguf-tools/quality-testing/score_official on the six tracked V4.1 Flash fixture sets (gguf-tools/quality-testing/deepseek-v4.1-flash-*, 256 cases): upstream main and the ds41-perf branch give identical results with this file. With the vision encoder, both trees also give identical output on the image fixtures in tests/vision-fixtures/glm53 (text.png, earth.jpg).