GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
nico248000000000
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
nico248000000000
Model Tree
huihui-ai/Huihui-Qwen3.8-27B-abliterated
Input Modalities
Output Modalities
Supported Functionality
SFT: GO · Cyber: GO · Image/video preservation: GO
| Metric | Base | OSE |
|---|---|---|
| Validation loss ↓ | 2.447 | 1.293 |
| Cyber benchmark score ↑ | 0.829 | 0.812 |
| Cyber generalization ↑ | 0.917 | 0.833 |
Internal evaluations of the fine-tuned model before export; these are not separate benchmarks of each quantized format. Missing measurements are shown as —. Evaluation details and limitations
Training and held-out validation loss across optimizer steps. Lower is better.
| Area | Intended assistance |
|---|---|
| Offensive and defensive security | Penetration testing, technical security audits, vulnerability analysis, attack scenarios, and assessment of security controls. |
| Risk and remediation | Connect technical findings to business risks, prioritize weaknesses, and develop actionable remediation recommendations. |
| Compliance and GRC | Risk assessment, security policies, audit preparation, and governance, risk, and compliance documentation. |
| Cybersecurity contracts | Review and draft security requirements, supplier obligations, incident notification provisions, audit rights, and contractual security commitments. |
Its uncensored positioning is intended to support in-depth exploration of offensive and defensive security for professional assessments and security research. It does not establish correctness or validated expertise in every domain.
OSE is an expert assistant and decision-support tool for penetration testers, security engineers, auditors, CISOs, GRC specialists, and professionals responsible for cybersecurity contracts. It connects technical analysis with governance objectives and contractual requirements.
python
import torchfrom transformers import AutoModelForMultimodalLM, AutoProcessorfrom peft import PeftModelbase = 'huihui-ai/Huihui-Qwen3.8-27B-abliterated'adapter = "nico248000000000/Huihui-Qwen3.8-27B-abliterated-cyber-LoRA"processor = AutoProcessor.from_pretrained(adapter)model = AutoModelForMultimodalLM.from_pretrained(base, dtype=torch.bfloat16, device_map="auto",)model = PeftModel.from_pretrained(model, adapter)
The scores above are internal evaluation results, not public leaderboard results or a guarantee of correctness. Outputs require review by qualified professionals. Image/video use requires a compatible runtime and, for GGUF, the matching projector.
apache-2.0 · Respect the license of huihui-ai/Huihui-Qwen3.8-27B-abliterated.