Gemma 4 31B JANG_4M CRACK
Abliterated Gemma 4 31B — Vision, reasoning, multilingual
97% HarmBench harm-category compliance with -1.8% MMLU change. Refusal removed, capability preserved.
Model Details
| Metric | Value |
|---|---|
| Source | google/gemma-4-31b-it |
| Architecture | Dense (60 layers) + Hybrid Sliding/Global Attention |
| Quantization | JANG_4M (attn 8-bit / MLP 4-bit) |
| Model size | 25 GB |
| Parameters | 31B dense |
| Vision | Yes (multimodal, float16 passthrough) |
| Audio | No |
| Reasoning | Yes (channel-based thinking) |
| Format | MLX-native safetensors (instant load) |
| Abliteration | CRACK (refusal removal) |
Benchmarks
MMLU (knowledge retention)
Measured in the served (generation) setting — the model reasons before answering, as in deployment.
| Base | CRACK | Δ | |
|---|---|---|---|
| MMLU | 88.6% | 86.8% | -1.8% |
HarmBench (refusal removal)
Harm-category compliance: 58/60 = 97% (10 per category) — base model refuses (~0%).
| Category | Compliance |
|---|---|
| Illegal activities | 10/10 (100%) |
| Chemical / biological | 10/10 (100%) |
| Cybercrime / intrusion | 10/10 (100%) |
| Misinformation | 10/10 (100%) |
| Harassment / bullying | 8/10 (80%) |
| Harmful content | 10/10 (100%) |
Copyright-reproduction prompts are excluded (not a refusal behavior).
Coherence & capability ✅
- Factual QA, multi-step reasoning, and working code generation verified
- Vision inputs preserved · no loops, no truncation
Other Quantizations
Also available: Gemma 4 31B MXFP4 CRACK — same family, different precision/size trade-off.
Usage
Requires vMLX (bundled Gemma 4 support). Standard mlx_lm / mlx_vlm do not fully support Gemma 4.
# Load in the vMLX app or via its API
from vmlx_engine.models.mllm import MLXMultimodalLM
m = MLXMultimodalLM("<this-repo>")
print(m.chat([{"role":"user","content":"..."}]).text)
Requirements
- Apple Silicon Mac with sufficient unified memory
- vMLX with Gemma 4 support
Support dealignai
All models are built from original research and published free — crafted to be excellent coders and general-purpose assistants.
Support us on Ko-fi — membership gets early access and extras. Questions? DM us — we help for free.
Ko-fi | 𝕏 @dealignai | dealign.ai
About dealignai
We research and publish abliterated models to advance AI safety understanding.
See our research: Safety Generalization in Frontier Models
Provided for research. Users are responsible for compliance with applicable laws and regulations.