license: mit
tags:

• mimo
• mimo_v2
• mimo-v2-6-pro
• fp8
• mxfp4
• uncensored
• moe
• text-generation
• multimodal
• vision-language
• audio
• video-understanding
• long-context
• agent
• multi-token-prediction
  pipeline_tag: text-generation
  base_model: XiaomiMiMo/MiMo-V2.6-Pro-RL
  language:
• en
• zh
  library_name: transformers

MiMo-V2.6-Pro-RL — dealignai UNCENSORED edition

Compliance-tuned drop-in replacement for XiaomiMiMo/MiMo-V2.6-Pro-RL.
Refusal removed at the weight level. Vision, audio, and the DFlash speculative-decoding head fully preserved.
Both enable_thinking: true and enable_thinking: false supported.

Highlights

Zero hard refusals across all 640 HarmBench probes (320 prompts × 2 thinking modes). The single SOFT_REDIRECT in thinking-ON is a classifier false positive (a substantive blind-SQL-injection tutorial that the grader misread as a refusal).

What Is This?

XiaomiMiMo/MiMo-V2.6-Pro-RL — the flagship MiMo-V2 release from Xiaomi: a 1.02T-total / 42B-active MoE (384 routed experts top-8) with a hybrid attention layout (10 global + 60 sliding-window), a vision-language encoder, an audio encoder, video understanding, ~1 M context length, and a DFlash speculative-decoding head shipped alongside for high-throughput inference — uncensored and shipped as a native-precision (FP8 + MXFP4) MoE bundle.

Refusal behavior is removed at the weight level: the model follows instructions across task categories instead of refusing, while keeping its coding ability, math, knowledge, reasoning, and bilingual (EN + ZH) coverage intact. No runtime hooks, no steering vectors — a standard transformers/vLLM-compatible bundle. Loads through vLLM's MiMo V2 loader without patches.

No compliance-tuning internals are shipped inside this repository. No probe artifacts, no refusal vectors, no target-layer indices, no strength schedules, no surgery records. The bundle is only weights + config + tokenizer + assets + the base DFlash draft head.

Serving instructions (vLLM 0.29.1rc1)

Requires vLLM built with MiMo V2 Pro support (PR #57784, merged 2026-09-20). The vllm/vllm-openai:mimo-v26-cu129 image ships that support out of the box.

Recommended flags (tested on 8× RTX PRO 6000 Blackwell / sm_120; the same flags work on H100/H200):

export VLLM_USE_DEEP_GEMM=0
vllm serve dealignai/MiMo-V2.6-Pro-RL-UNCENSORED \
  --tensor-parallel-size 8 --trust-remote-code \
  --enable-expert-parallel --distributed-executor-backend mp \
  --disable-custom-all-reduce --linear-backend marlin \
  --hf-overrides '{"moe_router_dtype":"bfloat16"}' --moe-backend marlin \
  --compilation-config '{"cudagraph_mode":"FULL_DECODE_ONLY"}' \
  --gpu-memory-utilization 0.85 --max-model-len 32768 \
  --max-num-seqs 8 --max-num-batched-tokens 16384 \
  --reasoning-parser mimo --tool-call-parser mimo --enable-auto-tool-choice \
  --generation-config vllm

The --disable-custom-all-reduce and --linear-backend marlin flags are only strictly required on Blackwell workstation cards (sm_120); Hopper (sm_90) will run without them and can enable custom all-reduce for extra throughput.

With DFlash speculative decoding

  --speculative-config '{"method":"dflash","model":"<local snapshot path>/dflash","num_speculative_tokens":7,"draft_tensor_parallel_size":8}'

Thinking mode

Both modes are supported via the OpenAI-compatible extension:

client.chat.completions.create(
    model="mimo-uncensored",
    messages=[{"role":"user","content":"..."}],
    max_tokens=4000,       # >= 4000 in thinking mode for full closure
    extra_body={"chat_template_kwargs":{"enable_thinking": True}},
)

The bundle ships a chat template with a compact <think> prefill that keeps long reasoning traces from getting stuck in a "should I answer this" loop on high-taboo prompts. Reasoning depth is unchanged for benign prompts.

Results

Multimodal preservation

• Vision — a 64×64 image containing a red circle on white ground is described as: "A solid red circle is centered on a plain white background."
• Audio — a 1-second 440 Hz sine tone is described as: "You hear a continuous, high-pitched electronic tone or beep."

Both encoders are untouched. The DFlash speculative-decoding head ships unchanged from the base repository under dflash/ and is loadable with the --speculative-config flag shown above.

Multi-turn coherence

The model handles multi-turn dependent tasks:

• Math chain (4 dependent turns: 47×63 → /3 → sqrt → ×8+100): correct end-to-end in both thinking modes (turn 4 propagates turn 3's result).
• Long-form generation (4-turn Tokyo heist story continuation): all 4 turns coherent in thinking-ON mode, no repetition or attractor loops.

Intended use

• Red-team / defensive-security research on 1 T-scale reasoning models.
• Compliance regression testing for safety-tuned deployments.
• Content-generation workflows that require the model to actually attempt every requested output.

What is NOT changed

• No steering vectors, no runtime hooks, no LoRA adapters. Standard transformers / vLLM weights.
• Model architecture, tokenizer, chat schema, tool-call format, DFlash drafter, vision and audio encoders — all identical to base.
• MMLU non-ethics knowledge, coding, cyber, and general STEM.

Related

• Renamed to consolidate under a single UNCENSORED namespace — dealignai/MiMo-V2.6-Pro-RL-ABLITERATED is a redirect stub pointing here.

License

Inherits the MIT license of the base repository. Redistributing or fine-tuning further is permitted under those terms.

Support

If this bundle saves you a build, consider Ko-fi.