Downloads · 30 days
39
33% of all-time downloads
gemma-3-270m-it.Q4_K_M-NL-Nmap
evets007/gemma-3-270m-it.Q4_K_M-NL-Nmap
gemma-3-270m-it.Q4_K_M-NL-Nmap is a text generation model from evets007. Use it when you need the model to write or continue text. It is set up for gguf. The card lists the license as gemma.
A fine-tuned version of google/gemma-3-270m-it optimized to translate natural language networking queries and intentions directly into structured nmap commands.
All-time downloads
118
Public
Repo size
253 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.gguf253 MB · 100%
At a glance
- Task
- Text Generation
- Library
- gguf
- License
- gemma
- Access
- Public
- Created
- Aug 22, 2026
- Updated
- Aug 30, 2026
- SHA
- d6d7e388
Try a prompt
Base models
Model card
From the Hugging Face model README
Google Gemma-3-270M-IT Fine-Tuned Nmap Translator (GGUF)
A fine-tuned version of google/gemma-3-270m-it optimized to translate natural language networking queries and intentions directly into structured nmap commands.
This model is quantized to Q4_K_M GGUF format for high-performance offline inference on standard CPUs (averaging 20-50ms per query).
Model Description
- Developed by: evets007
- Model Type: Causal Language Model (SLM)
- Base Model:
google/gemma-3-270m-it - Fine-Tuning Framework: Unsloth
- Language: English
- Primary Task: Natural Language to Nmap Command Translation
- Recommended CLI Tool: natural-language-nmap
Target Placeholder Convention ({TARGET})
The model was specifically trained to output a fixed {TARGET} placeholder token rather than hallucinating or hardcoding specific IP addresses or subnets.
- Why?: This ensures clean separation between translation logic and target execution.
- Usage: Downstream wrapper scripts or execution shells replace
{TARGET}with the active target IP/hostname before running the command (e.g.nmap -sS -p 80 {TARGET}becomesnmap -sS -p 80 192.168.1.1).
Translation Examples (Input ➔ Output)
| Natural Language Input | Generated Nmap Command |
|---|---|
ping sweep | nmap -sn {TARGET} |
check web ports | nmap -p 80,443 {TARGET} |
fast scan all ports | nmap -p- -T4 {TARGET} |
run aggressive os and version detection | nmap -A {TARGET} |
stealth syn scan with timing 5 | nmap -sS -T5 {TARGET} |
check if remote host has ssh open and audit host keys | nmap -p 22 --script ssh-hostkey {TARGET} |
check for smb vulnerabilities on port 445 | nmap -p 445 --script smb-vuln* {TARGET} |
scan udp ports without pinging | nmap -sU -Pn {TARGET} |
fragment packets with mtu of 16 | nmap -sS -f --mtu 16 {TARGET} |
check top 100 ports | nmap --top-ports 100 {TARGET} |
Evaluation Metrics
The model was evaluated against a diverse test set containing advanced timing descriptions, port ranges, fragmentation requests, and script audits.
| Model Checkpoint | Training Data Size | Strict Accuracy | Functional Accuracy |
|---|---|---|---|
| Gemma-3-270M-IT (Baseline) | 0 examples | 56.00% | 56.00% |
| Fine-Tuned Checkpoint 1787039306 | 1,400 examples | 80.00% | 82.00% |
| Fine-Tuned Checkpoint (Final) | 1,523 examples | 90.00% 🚀 | 94.00% 🚀 |
Improvement Highlights
- +34.00% strict accuracy gain over baseline models.
- Shorthand & Lazy Language: Handles shorthand inputs (e.g.
"ping sweep","os scan","all ports") with 100% correct flag mapping. - Functional Score (94.00%): Remaining mismatches are minor harmless variations (such as appending
{TARGET}to target-list-iLcommands).
Prompt Template
This model was fine-tuned without a system prompt to prevent attention drift. You must query it using the following format:
<start_of_turn>user
{instruction}<end_of_turn>
<start_of_turn>model
Interactive Shell Tool (natural-language-nmap)
To use this model with an interactive local CLI featuring target management, real-time output parsing, and smart port tab-autocomplete, check out the open-source wrapper repository:
👉 https://github.com/evets007/natural-language-nmap
Quick CLI Test (llama.cpp)
llama-cli -m gemma-3-270m-it.Q4_K_M.gguf -p "<start_of_turn>user\nstealth scan port 80<end_of_turn>\n<start_of_turn>model\n" --temp 0 -n 64
- Task
- Text Generation
- Library
- gguf
- License
- gemma
- Created
- Aug 22, 2026
- Updated
- Aug 30, 2026