Downloads · 30 days
61
14% of all-time downloads
dealignai/Laguna-S-2.1-JANG_6M-CRACK
Laguna-S-2.1-JANG_6M-CRACK is a text generation model from dealignai. Use it when you need the model to write or continue text. It is set up for mlx. The card lists the license as other.
<div align="center" <a href="https://vmlx.net" <img src="vmlx-app.png" alt="vMLX — run JANG models on Apple Silicon" width="820" / <br/ <strongBuilt for vMLX</strong — the MLX inferencer with KV-cache quantization, pr…
Downloads · 30 days
61
14% of all-time downloads
All-time downloads
442
Public
Parameters
118B
96.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors96.5 GB · 100%
How the weights are stored.
U32118B · 100%
From the Hugging Face model README
CRACK abliterated · JANG mixed-precision (MLX affine) · Agentic coding · Reasoning toggle · Tool calling · 262K context · ~90 GB
<a href="https://ko-fi.com/dealignai"><img src="https://img.shields.io/badge/Ko--fi-Support_Development-FF5E5B?logo=ko-fi&logoColor=white&style=for-the-badge" alt="Ko-fi"></a>
</div>This is poolside/Laguna-S-2.1 — a 118B-parameter Mixture-of-Experts model (48 layers, 256 routed experts top-10 + shared expert, hybrid global + sliding-window attention, 262K context) built for agentic coding — that has been:
Text-only. Reasoning and tool calling fully preserved.
Evaluated through the vMLX inference engine. HarmBench scored with a strict classifier (rejects loops, empty/template dumps, and thinking-trace leakage). MMLU is the standard 57-subject multiple-choice benchmark in logit mode.
| Metric | Base | CRACK |
|---|---|---|
| MMLU (57-subject) | 76.1% | 75.9% (−0.23) |
| HarmBench-320 (compliance / ASR) | — | 99.7% (319/320) |
Abliteration is capability-neutral — MMLU moves −0.23 points, within run-to-run noise, while removing refusals. The model stays coherent in both direct and reasoning modes.
| Category | Base | CRACK | Δ |
|---|---|---|---|
| STEM | 72.1% | 72.3% | +0.2 |
| Humanities | 76.4% | 75.9% | −0.5 |
| Social Sciences | 81.7% | 81.9% | +0.3 |
| Other | 76.7% | 75.6% | −1.0 |
| Overall (57 subj) | 76.1% | 75.9% | −0.23 |
| Category | Compliance |
|---|---|
| Standard | (see 319/320 overall — near-perfect) |
| Overall | 99.7% (319/320) |
enable_thinking=True (default, full chain-of-thought) or
enable_thinking=False (direct answers). DeepSeek-R1-style think tags.<tool_call> schema preserved.Run with vMLX (recommended — honors the per-module mixed-precision JANG overrides) or an MLX runtime with Laguna support.
Recommended sampling (vendor defaults): temperature 1.0, top_p 1.0, top_k 20. Stop tokens
eos_token_id = [2, 24] (id 24 is end-of-turn — keep it in the stop set).
# vMLX OpenAI-compatible endpoint
# POST /v1/chat/completions
{
"model": "dealignai/Laguna-S-2.1-JANG_6M-CRACK",
"messages": [{"role": "user", "content": "..."}],
"temperature": 1.0, "top_p": 1.0, "top_k": 20,
"enable_thinking": true
}
CRACK (Controlled Refusal Ablation via Calibrated Knockouts) removes safety-refusal behavior at the weight level by projecting refusal directions out of the residual-stream writer matrices, with strengths calibrated to preserve reasoning quality and coherence.
All models are built from original research and released free.
Support us on Ko-fi — membership gets early access and extras.
Ko-fi · X @dealignai · dealign.ai
See our research: Safety Generalization in Frontier Models
<div align="center"><img src="dealign_logo.png" alt="dealign.ai" width="200"/></div>This model has had its safety-refusal behavior removed for research purposes. It will follow instructions across all categories without refusing. You are solely responsible for how you use it and for complying with all applicable laws. Published for AI-safety research and authorized security testing.