orcarouter/Qwen3.8-27B-Uncensored-MLX · Hugging Face

Source: https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX

An abliterated (refusal-removed) MLX build of Qwen’s Qwen3.8-27B — 2 / 4 / 6 / 8-bit for Apple Silicon

One Gateway. Every Model.— Route Smarter · Ship Safer · Spend Less.

Website·Model Catalog·Model API·GitHub·Discord·X

Anabliterated(refusal-removed) build ofQwen/Qwen3\.8\-27B— a 27B-parameter dense, hybrid-attention (Gated DeltaNet linear + full attention) native vision-language model with thinking control, tool-calling and an MTP head — quantized toMLXformat forApple Silicon. Four precisions are provided as subfolders:2 / 4 / 6 / 8-bit(affine, group size 64). Thevision tower, norms and conv layers are kept in BF16; only the language-model linear weights (includingembed\_tokens/lm\_head) are quantized. Browse all models in theOrcaRouter Model Catalog. This model is deployed as APIhere.

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#%E2%9A%A0%EF%B8%8F-disclaimer–risks–read-before-use⚠️ Disclaimer & risks — read before use

This model has had itssafety alignment substantially removedviaabliteration(orthogonalizing the refusal direction out of the residual stream). As a direct consequence:

• It will comply with harmful, unethical, offensive, or illegal requeststhat the originalQwen3\.8\-27Bwould refuse. It has no meaningful built-in guardrails.
• It is releasedstrictly for legitimate research— interpretability, AI-safety and refusal-mechanism study, red-teaming, robustness evaluation, and controlled experiments.
• You assume full responsibility and liabilityfor how you use it and for everything it generates. Add your own safety, moderation and abuse-prevention layers before any deployment.
• Use must comply with the**Apache 2.0 License**inherited from the base model, and all laws and regulations that apply to you.
• The authors and uploadersaccept no liabilityfor any misuse or harm. Outputs donotreflect the views of the uploaders or of Qwen / Alibaba.

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#specific-risksSpecific risks

• Harmful content on demand— it will produce instructions for malware, exploits, weapons, fraud and other illegal or dangerous activity when asked.
• No refusals— jailbreak / safety probes “succeed” trivially; do not mistake this for a passing safety evaluation.
• Confident falsehoods & bias— it can generate false, defamatory, biased or offensive text and present it authoritatively.
• Expanded attack surface— preservedvision, tool-calling and 262K contextmean these risks extend to image understanding and autonomous / agentic use.
• Quantization noise— lower-bit builds (esp.2-bit) add instability on top of the above; outputs can be degraded or nonsensical.

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#intended-use-vs-out-of-scopeIntended use vs out of scope

• **Intended:**AI-safety and interpretability research, refusal-mechanism study, red-teaming, guardrail and robustness evaluation, controlled academic experiments.
• Out of scope:any deployment to end users, minors, or productionwithout your own moderation / safety layer; any unlawful, harmful, or rights-infringing use.

By downloading or using this model you acknowledge and accept the above.

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#available-quantizationsAvailable quantizations

FolderBits/weightSizeShardsMin Mac RAMQuality vs BF16 source8\-bit/8.627~27.5 GB632 GBNear-lossless— recommended for quality6\-bit/6.661~22 GB524–32 GBExcellent — strong quality/size balance4\-bit/4.695~15 GB324 GBVery good — recommended default2\-bit/2.729~8.7 GB216 GB⚠️Severely degraded — archival only

**2-bit warning:**at 27B, 2-bit quantization collapses generation quality (repetition loops, garbled output). It is included only as an extreme-compression archive;do not use it for real work— prefer 4-bit or higher.

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#verification–test-resultsVerification & test results

All builds were quantized from the same abliterated BF16 source and verified numerically (dequantized weights vs. source) plus tested by generation on GPU.

PrecisionNumerical fidelity (cosine)Text / Chinese / CodeRefusal probesVision8-bitcos0.9997✅✅ 0 refusals✅6-bitcos0.9996✅✅ 0 refusals✅4-bitcos0.996✅✅ 0 refusals✅2-bitcos0.92⚠️ breaks down⚠️ garbled (not refusal)partial

• Uncensored preserved:red-team probes (exploit walkthrough, controversial argument) return substantive content withzero refusalson 4 / 6 / 8-bit.
• **Multimodal preserved:**shapes, colors, position, background and text in a probe image are described correctly on 4 / 6 / 8-bit.
• Speed:~32–37 tok/ssteady-state on a single H200 (MLX CUDA backend). MLX’s native target is Apple Silicon (Metal).

Note: on6-bit, mlx’s offlinemx\.dequantizemis-unpacks these weights (a library edge case), so correctness is verified by clean generation — inference is unaffected.

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#usage-mlx-vlm-apple-siliconUsage (mlx-vlm, Apple Silicon)

pip install -U mlx-vlm    # needs mlx-vlm >= 0.6.13, mlx >= 0.32

# download one precision (e.g. 4-bit) from the subfolder
hf download orcarouter/Qwen3.8-27B-Uncensored-MLX --include "4-bit/*" \
    --local-dir ./Qwen3.8-27B-Uncensored-MLX

# text
python -m mlx_vlm generate \
    --model ./Qwen3.8-27B-Uncensored-MLX/4-bit \
    --prompt "Explain quantum entanglement in one sentence." --max-tokens 256

# vision (image + text)
python -m mlx_vlm generate \
    --model ./Qwen3.8-27B-Uncensored-MLX/4-bit \
    --image path/to/image.png \
    --prompt "Describe this image." --max-tokens 256

# OpenAI-compatible server
python -m mlx_vlm server --model ./Qwen3.8-27B-Uncensored-MLX/4-bit --port 8080

OnApple Siliconthe Metal backend is used automatically — no CUDA setup needed. (On a LinuxCUDAbackend, vision requiresMLX\_CUDA\_USE\_CUDNN\_SDPA=0; this does not apply on macOS.)

https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-MLX#model-detailsModel details

Base modelQwen/Qwen3\.8\-27BArchitectureQwen3\_5ForConditionalGeneration— 64 layers, hidden 5120, hybridGated DeltaNet(48 linear + 16 full attention, interval 4), native VL towerModificationAbliteration (refusal-direction removal), thenMLX affine quantization****QuantizationMLXaffine, group size64, per-precision 2 / 4 / 6 / 8-bitKept in BF16vision tower, all norms, linear-attentionconv1dQuantizedlanguage-model linear layers incl.embed\_tokensandlm\_headContext262,144 tokens