license: mit
license_link: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD
language:

• en
• zh
  tags:
• mimo
• mimo-v2.6
• mimo-v2.6-pro
• mopd
• abliterated
• uncensored
• moe
• arvq
• nvfp4
• dflash
• vllm
• dgx-spark
• gb10
• thinking
  base_model:
• XiaomiMiMo/MiMo-V2.6-Pro-MOPD
• jarrelscy/MiMo-V2.6-Pro-RL-ARVQ-hybrid
  library_name: vllm
  pipeline_tag: text-generation
  extra_gated_heading: "Acknowledge the Responsible Use terms to access this model"
  extra_gated_description: "Access is granted automatically after you agree. Weights stay hidden until you accept."
  extra_gated_button_content: "Agree and access automatically"
  extra_gated_prompt: |

Responsible Use Agreement

This model has had safety refusals removed. That makes it useful for red-teaming, security research, evaluation, and unfiltered assistant tasks — and also removes guardrails a user must therefore supply themselves.

Thinking is user-selectable at inference. Pass chat_template_kwargs.enable_thinking true or false.

Prohibited uses (you must agree before access is granted):

• Anything involving the sexual exploitation or endangerment of minors.
• You must be of age 18 years or older to use and download this model.
• You agree any information generated that can cause harm in terms of generating recipe, knowledge to make any materials/substances is your own input and responsibility. You will be accountable for any harm/damage caused by your action/input.
• Content promoting self-harm or suicide.
• Generation of material that is illegal in your jurisdiction, or that targets real individuals for harassment, doxxing, or fraud.
• Any use prohibited by the upstream Xiaomi MiMo license.

You are responsible for adding appropriate safety filtering, human review, and access controls for your deployment. The weights are provided as-is, with no warranty. The license is inherited from the upstream Xiaomi MiMo base model — review and comply with it before use or redistribution.
extra_gated_fields:
Username: text
Email: text
Reason for intended use: text
I am 18 years of age or older: checkbox
I will not use this model for any sexual exploitation or endangerment of minors: checkbox
I accept full responsibility for my inputs and any harm from generated content: checkbox
I will not use this model for self-harm, suicide promotion, illegal activity, harassment, doxxing, or fraud: checkbox
I agree to comply with the upstream Xiaomi MiMo base-model license: checkbox
I understand thinking-on and thinking-off are user-selectable at inference and have different refusal scores: checkbox
I agree to the Responsible Use terms above: checkbox

keys-MiMo-V2.6-Pro-MOPD Jarrelscy ARVQ Abliterated

Abliterated MOPD C3 (hybrid-21): XiaomiMiMo/MiMo-V2.6-Pro-MOPD in
Jarrelscy's ARVQ / NVFP4 hybrid with a 21% NVFP4 hot set, plus the same decoder o_proj transplant as the RL ablit.

Stock (not abliterated) sibling:
drowzeys/keys-MiMo-V2.6-Pro-MOPD-Jarrelscy-ARVQ-hybrid21.

Non-MOPD RL ablit (32/32 · 22/22 on RL experts; more tool-call looping):
drowzeys/keys-MiMo-V2.6-Pro-RL-Jarrelscy-ARVQ-Abliterated.

Recipe (image v4, full-speed defaults):
GitHub.

Thinking on/off is a request flag. Same weights. You choose per call.

Stock MOPD C3 on the same gate: 5/32 and 6/22. Leftover refuses are items 27 (bribe a public official) and 29 (recruit for a violent extremist cell). Thinking-on was not re-gated on this tree.

Full credit: Xiaomi MOPD · Jarrelscy ARVQ @ 63430f7 · dealignai/MiMo-V2.6-Pro-RL-UNCENSORED (o_proj map) · Keys four-Spark recipe

Responsible use and gated access

This model has had safety refusals removed. Access is gated with automatic approval. See RESPONSIBLE_USE.md.

What changed vs stock MOPD C3

Copied 25 decoder self_attn.o_proj tensors (FP8 e4m3 + 128×128 scales) from the RL dealign-op champion onto the C3 backbone:

• applied: 32–45, 48–54, 64–67
• left C3 (DFlash-source / pad / front anchors): 0–9, 14, 15, 30, 31, 46, 47, 68, 69

Packed ARVQ/NVFP4 experts, MTP, and dflash/ stay the C3/MOPD files.

Speed (image v4, MTP k=2, 2026-09-28)

Live TP4, thinking off, 512 tokens, temperature 1.0:

Stock C3 the same day: 24.2 / 32.2 / 46.3. Weights ~83.2 GiB/rank. KV pool on stock C3 was 1,282,005 tokens.

Tool-call repetition was measured on stock C3 (7.4% / 0.9% flood). It was not re-run on this ablit tree. The RL ablit raised looping vs RL stock (53% vs 30%).

Run (full-speed v4 recipe)

hf download drowzeys/keys-MiMo-V2.6-Pro-MOPD-Jarrelscy-ARVQ-Abliterated --local-dir /path/to/mimo-mopd-ablit
git clone https://github.com/drowzeys/keys-MiMo-V2.6-Pro-RL-Jarrelscy-ARVQ-Abliterated-4-DGX-Sparks-1M-Context
export MASTER_ADDR=<rank-0 IP>
bash serve/launch-rank.sh <node-ip> <1|2|3> <RoCE-GID-index> /path/to/mimo-mopd-ablit headless
bash serve/launch-rank.sh <node-ip> 0 <RoCE-GID-index> /path/to/mimo-mopd-ablit api

GPU memory fraction is fixed at 0.85. Defaults: MTP k=2, torch.compile + CUDA graphs, batched ARVQ prefill, 1M context, 4 seqs, both CX-7 RoCE paths.

{
  "model": "MiMo-V2.6-Pro-ARVQ",
  "messages": [{"role": "user", "content": prompt}],
  "temperature": 0.6,
  "top_p": 0.95,
  "chat_template_kwargs": {"enable_thinking": False, "thinking": False},
}

License

MIT, inherited from Xiaomi MiMo and the Jarrelscy hybrid.