---

language:

• en
• ar

tags:

• cybersecurity
• qwen2.5
• gguf
• saif

license: apache-2.0

base_model: BinSaqban/Averroes-Q-Instruct

---

Saif-Kernel-7B 🗡️

Part of the SAIF (سيف) expert suite — Arabic-named security AI models.

File

| File | Size | Format |

|------|------|--------|

| Saif-Kernel-7B-v0.5-Q4_K_M.gguf | ~4.36 GB | Q4_K_M (llama.cpp compatible) |

Usage

Ollama

# Download and create
curl -L -o Saif-Kernel-7B-Q4.gguf "https://huggingface.co/BinSaqban/Saif-Kernel-7B-GGUF/resolve/main/Saif-Kernel-7B-v0.5-Q4_K_M.gguf"

# Create Modelfile
cat > Modelfile << 'EOF'
FROM Saif-Kernel-7B-Q4.gguf
PARAMETER temperature 0.3
PARAMETER num_ctx 4096
SYSTEM "You are Saif, a cybersecurity analysis assistant."
EOF

ollama create saif-kernel -f Modelfile
ollama run saif-kernel

llama.cpp / LM Studio

Just open the .gguf file directly — it's a standard Q4_K_M quantization.

Hardware Requirements

| RAM | Performance |

|:---:|:----------:|

| 8 GB | ✅ Works with num_ctx 2048, may use swap |

| 16 GB | ✅ Full speed, num_ctx 8192 |

| 32 GB+ | ✅ Max context |

About SAIF Suite

| Model | Purpose |

|-------|---------|

| Saif-Kernel-7B | 🧠 Core reasoning & cybersecurity analysis |

| Saif-Browser-7B | 🌐 Browser automation & web interaction |

| Saif-Router-7B | 🔀 Query classification & routing |

Built on Averroes-Q-Instruct — Qwen2.5 fine-tuned for Arabic-English bilingual security applications.