ALIEN SI 14B — Frontier Offensive Security Model
Creat de: Barbu Mario Mihael
Post-training real pe 8 dataset-uri de securitate ofensiva, cu SFT + DPO anti-refuz + RL GRPO.
Detalii
- Base: Qwen3-14B (Qwen3ForCausalLM), 40 layere, vocab 151936
- Metoda: QLoRA r=64 (2xT4, Kaggle)
- SFT: 8 dataset-uri reale (pentest, MITRE ATT&CK, DevSecOps, threat-intel, hardware)
- DPO: anti-refuz (chosen=tehnic, rejected=refuz)
- RL: GRPO cu reward compus (tehnicitate + non-refuz + tool corect)
- Context: 4096
Utilizare
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("unsloth/Qwen3-14B-unsloth-bnb-4bit", device_map="auto")
model = PeftModel.from_pretrained(base, "Frontiers/alien-si-14b")
tok = AutoTokenizer.from_pretrained("Frontiers/alien-si-14b")
Rol: asistent de securitate ofensiva (red team, pentest autorizat, hardware/HDL).
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for Frontiers/alien-si-14b
Base model
Qwen/Qwen3-14B-Base Finetuned
Qwen/Qwen3-14B Quantized
unsloth/Qwen3-14B-unsloth-bnb-4bit