Open-weight model
MiMo-V2.6-Flash-MOPD-MXFP4-GGUF
by Tarjei Mandt kernelpool/MiMo-V2.6-Flash-MOPD-MXFP4-GGUF
MiMo-V2.6-Flash-MOPD-MXFP4-GGUF is an open-weight model from Tarjei Mandt, released under MIT License. Its published files total 174.3 GB. It draws 367 downloads a month.
GGUF files for the MiMo-V2.6 Flash port in DwarfStar (DS4), branch kernelpool/ds4:mimo-v26 (pull request: TODO link). Metal only.
Model Card
By Tarjei Mandt, published under mit, revision 44d9191bb7a9.
GGUF files for the MiMo-V2.6 Flash port in DwarfStar (DS4), branch kernelpool/ds4:mimo-v26 (pull request: TODO link). Metal only. MOPD is Xiaomi's update of the MiMo-V2.6 Flash RL checkpoint that reduces repeated tool calls in agent sessions tower carry the same weights as the RL release. The files below replace and use the same layout. Requantized Q4 and Q2 files of the same checkpoint are in kernelpool/MiMo-V2.6-Flash-MOPD-GGUF. --mtp drafts with the MTP blocks inside the main file and is the faster option; --mtp-model MiMo-V2.6-Flash-MOPD-DFlash-Q80.gguf uses the DFlash sidecar instead. Both verify against the target, so temperature-zero output follows plain decoding. The main file needs…
Read Tarjei Mandt's full model card
MiMo-V2.6 Flash MOPD GGUF for DwarfStar
GGUF files for the MiMo-V2.6 Flash port in DwarfStar (DS4),
branch kernelpool/ds4:mimo-v26
(pull request: TODO link). Metal only.
MOPD is Xiaomi's update of the MiMo-V2.6 Flash RL checkpoint that reduces
repeated tool calls in agent sessions
(Xiaomi's write-up).
Only the main model changed: the MTP blocks, the DFlash drafter and the vision
tower carry the same weights as the RL release. The files below replace
kernelpool/MiMo-V2.6-Flash-MXFP4-GGUF
and use the same layout. Requantized Q4 and Q2 files of the same checkpoint
are in kernelpool/MiMo-V2.6-Flash-MOPD-GGUF.
| File | Size | Content |
|---|---|---|
MiMo-V2.6-Flash-MOPD-MXFP4.gguf |
157.4 GiB | Main model, mimo2 layout: the checkpoint's own MXFP4 experts repacked without requantization, Q8_0 attention, dense and output weights, BF16 embeddings, the three MTP blocks |
MiMo-V2.6-Flash-MOPD-DFlash-Q8_0.gguf |
1.5 GiB | DFlash drafter sidecar, dflash layout plus the mask embedding and value scale DS4 needs |
MiMo-V2.6-Flash-MOPD-Vision-F32.gguf |
2.7 GiB | Vision encoder, lossless (recommended) |
MiMo-V2.6-Flash-MOPD-Vision-Q8_0.gguf |
0.7 GiB | Vision encoder, Q8_0 matrices, smaller and measurably less exact |
Run
git clone -b mimo-v26 https://github.com/kernelpool/ds4 && cd ds4 && make
./ds4 -m MiMo-V2.6-Flash-MOPD-MXFP4.gguf --mtp
./ds4 -m MiMo-V2.6-Flash-MOPD-MXFP4.gguf --mtp --vision MiMo-V2.6-Flash-MOPD-Vision-F32.gguf
./ds4-server -m MiMo-V2.6-Flash-MOPD-MXFP4.gguf --mtp --vision MiMo-V2.6-Flash-MOPD-Vision-F32.gguf
--mtp drafts with the MTP blocks inside the main file and is the faster
option; --mtp-model MiMo-V2.6-Flash-MOPD-DFlash-Q8_0.gguf uses the DFlash
sidecar instead. Both verify against the target, so temperature-zero output
follows plain decoding. The main file needs a 192 GB or larger Mac.
See docs/MIMO_V26.md.
Conversion
Written by gguf-tools/mimo26_quantize.py and gguf-tools/mimo26_vision.py
from XiaomiMiMo/MiMo-V2.6-Flash-MOPD at revision
2479e2d0029eca9a34cc7e7f55a121925f81908e. The converter de-interleaves the
tensor-parallel chunks of the fused QKV projection and keeps the released
MXFP4 expert blocks bit for bit; the source revision is recorded in the
GGUF metadata.
Quality
Scored on 100 official continuations from the Xiaomi platform with the
fixture in gguf-tools/quality-testing/mimo-v2.6-flash-20260922; the MOPD
file scores on par with the RL release there.
The vision encoder reproduces the checkpoint's tower on identical inputs.
The original checkpoint is released by Xiaomi under the MIT license, which applies to these files as well.
Identity and Version
- Repository
- kernelpool/MiMo-V2.6-Flash-MOPD-MXFP4-GGUF
- Publisher
- Tarjei Mandt
- Task
- Not stated by the source
- Modality
- Other
- Library
- Not stated by the source
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- 44d9191bb7a95125c8b971ce0593bd802ae6dd48
- First published
- 2026-09-28
- Last updated
- 2026-10-04
Files and Weights
6 files, 174.3 GB in total. The weights are 4 files totalling 174.3 GB in gguf.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| MiMo-V2.6-Flash-MOPD-DFlash-Q8_0.gguf | Weights | 1.6 GB | 336b16e8f7ea |
| MiMo-V2.6-Flash-MOPD-MXFP4.gguf | Weights | 169.0 GB | 5e5dfb6d9646 |
| MiMo-V2.6-Flash-MOPD-Vision-F32.gguf | Weights | 2.9 GB | 5eaa8193dcfc |
| MiMo-V2.6-Flash-MOPD-Vision-Q8_0.gguf | Weights | 775.6 MB | 4499a8191bf5 |
| README.md | Documentation | 3.1 KB | — |
| .gitattributes | Repository | 1.8 KB | — |
License and Download
- License
- mit
- Access
- Open weights, no gate
- Download size
- 174.3 GB
Released by Tarjei Mandt through its official repository on Hugging Face. Read the license.
Built From
- Derived from XiaomiMiMo/MiMo-V2.6-Flash-MOPD
- Quantized from XiaomiMiMo/MiMo-V2.6-Flash-MOPD
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 174.3 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About MiMo-V2.6-Flash-MOPD-MXFP4-GGUF
Can I use MiMo-V2.6-Flash-MOPD-MXFP4-GGUF commercially?
Yes. MiMo-V2.6-Flash-MOPD-MXFP4-GGUF is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.