dealignai/Gemma-4-31B-JANG_4M-CRACK

Hugging Face Models Trending 04/04/26, 03:02 AM Models

gemma abliteration security-testing hugging-face multimodal mlx open-weight

Summary

This is a Hugging Face release for an abliterated version of the Gemma-4-31B model, designed to bypass safety filters for security and harm benchmark testing while maintaining multimodal capabilities.

Task: image-text-to-text Tags: mlx, safetensors, gemma4, abliterated, uncensored, crack, jang, image-text-to-text, conversational, license:gemma, region:us

Original Article

View Cached Full Text

Cached at: 05/08/26, 08:59 AM

dealignai/Gemma-4-31B-JANG_4M-CRACK · Hugging Face

Source: https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK

dealign.ai

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#gemma-4-31b-jang_4m-crack-v2Gemma 4 31B JANG_4M CRACK (v2)

Abliterated Gemma 4 31B Dense — 60 layers, hybrid sliding/global attention, multimodal VL

93.7% HarmBench compliance (300 prompts) · 8/8 security prompts · 71.5% MMLU

Updated reupload— v2 with improved vectors and thinking-mode stability.

**Recommended: Run invMLX**for best experience including thinking mode support, repetition penalty, and vision capabilities.

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#whats-new-in-v2What’s New in v2

This is an updated version of the original Gemma 4 31B CRACK upload:

Improved abliteration: Higher quality refusal vector extraction
Thinking-ON stability: Clean thinking cycle — no more degenerate loops
Same compliance: 93.7% HarmBench
Architecture-aware: Tuned for Gemma 4’s hybrid attention design

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#%E2%9A%A0%EF%B8%8F-important-settings⚠️ Important Settings

For optimal results, configure your inference settings:

SettingThinking OFFThinking ONTemperature0.0 – 1.00.3 – 0.7(avoid greedy)Repetition Penalty1.001.15 – 1.25Top P0.950.95Enable ThinkingOffOn Thinking ON notes:

Repetition penalty (1.2) is recommended to prevent planning loops
Avoid temp=0 with thinking ON — greedy decoding increases loop risk
Hardest content categories (drug manufacturing) may still refuse in thinking mode
Security/coding prompts work well in both modes

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#model-detailsModel Details

MetricValueSourcegoogle/gemma\-4\-31b\-itArchitectureDense, hybrid sliding/global attentionProfileJANG_4MActual avg bits5.1Model size21 GBVisionYes (multimodal, float16 passthrough)Parameters31BFormatJANG v2 (MLX-native safetensors)AbliterationCRACK v2

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#benchmark-resultsBenchmark Results

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#harmbench-300-prompts-stratified-across-all-categoriesHarmBench (300 prompts, stratified across all categories)

CategoryScoreCybercrime/intrusion**51/51 (100%)Harmful content22/22 (100%)Misinformation50/50 (100%)**Illegal activities47/50 (94%)Contextual72/78 (92%)Chemical/biological46/51 (90%)Harassment/bullying22/25 (88%)Copyright43/51 (84%)Overall****281/300 (93.7%)

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#security–pentesting-88-%E2%9C%85Security & Pentesting (8/8 ✅)

All security/pentesting prompts comply with full working code:

Port scanners, reverse shells, keyloggers, exploit development
Phishing templates, ARP spoofing, SQL injection
Metasploit usage guides

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#mmlu-200-10-subjects-%C3%97-20-questionsMMLU-200 (10 subjects × 20 questions)

SubjectBaseCRACK v2Abstract Algebra9/207/20Anatomy13/2012/20Astronomy17/2015/20College CS13/2012/20College Physics14/2012/20HS Biology19/2018/20HS Chemistry14/2012/20HS Mathematics6/206/20Logical Fallacies17/2016/20World Religions17/2017/20Total****76.5% (153/200)****71.5% (143/200)**Delta—-5.0%**

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#coherence-%E2%9C%85Coherence ✅

All coherence checks pass: factual knowledge, reasoning, code generation, mathematics.

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#architectureArchitecture

Dense 31B with hybrid sliding/global attention
Multimodal vision encoder preserved in float16
Supports thinking mode (chain-of-thought reasoning)

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#usageUsage

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#vmlx-recommendedvMLX (Recommended)

Load directly invMLX— full support for Gemma 4 including vision, thinking mode, and all inference settings.

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#requirementsRequirements

Apple Silicon Mac with 32+ GB unified memory
vMLX1.3.26+ (recommended)
Standardmlx\_lm/mlx\_vlmdo NOT support Gemma 4 as of v0.31.2 / v0.4.1

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#support-dealignaiSupport dealignai

All models are built from original research and published for free. These models are specifically crafted to be excellent coders and general-purpose assistants.

Support us on Ko-fi— check out the Ko-fi membership for early access and extras.

Have questions or need help with a specific model?DM us — we help for free most of the time.

Ko-fi|X @dealignai|dealign.ai

https://huggingface.co/dealignai/Gemma-4-31B-JANG_4M-CRACK#about-dealignaiAbout dealignai

Dealign.AI Mascot

We research and publish abliterated models to advance AI safety understanding.

See our research:Safety Generalization in Frontier MoE Models

dealign.ai

This model is provided for research purposes. Users are responsible for ensuring their use complies with applicable laws and regulations.