Cluster Worker v2 is a highly specialized Small Language Model (SLM) fine-tuned on 65,000+ Linux & DevOps task samples. It is engineered specifically to act as a deterministic, zero-fluff CLI execution worker inside dual-agent AI architectures.
Unlike standard chat models, this model uses response-only loss masking to completely eliminate conversational filler, explanations, and unnecessary Markdown wrappers — returning strictly valid, executable Bash commands.
Licensing & Commercial Use
This model is released under the Business Source License 1.1 (BSL 1.1).
Free Use: Free for non-production, testing, academic, and evaluation purposes.
Commercial Use: If you wish to use this model in production for commercial purposes, applications, or enterprise services, you must acquire a commercial license.
Contact for Enterprise Licensing
For production keys, custom fine-tuning on your company data, or commercial licensing inquiries, please reach out directly:
Zero Conversational Fluff: No "Sure, here is your command:" or introductory text. It outputs raw code meant directly for stdout/terminal execution.
Enterprise Sysadmin Coverage: Fine-tuned on multi-step Linux administrative tasks, file system manipulation, log parsing (awk, grep, sed), networking, and systemd management.
Ultra-Low Footprint: Built on Qwen2.5-Coder-0.5B-Instruct and quantized to Q8_0, taking up under 600 MB of VRAM/RAM.
Optimized for Dual-Agent Architectures: Serves as a zero-cloud-token execution node when paired with primary reasoning models (e.g., Gemini, Claude, GPT-4o, or a 7B Architect model).
Quickstart & Deployment
Method 1: Automated Script (Recommended)
If using Debian/Ubuntu Linux, run the official automated setup script to install Ollama, build the local wrapper, and launch the worker:
You can run this model directly in Ollama using the ChatML standard format.
Create a file named Modelfile:
dockerfile
1FROM hf.co/zk-mohammad/cluster-worker-v2-0.5b-GGUF23TEMPLATE """<|im_start|>system
4You are a strict, zero-fluff Linux terminal worker. Return strictly the bash command, no explanations, no markdown chat, no formatting.<|im_end|>
5<|im_start|>user
6{{ .Prompt }}<|im_end|>
7<|im_start|>assistant
8"""
910PARAMETER temperature 0.0
11PARAMETER top_k 1
12PARAMETER stop "<|im_end|>"
13PARAMETER stop "<|im_start|>"
Build and run the model:
bash
1ollama create worker -f Modelfile
2ollama run worker "Find all files ending in .log in /var/log older than 7 days and delete them."
Method 3: llama.cpp CLI
Run directly via llama-cli with Jinja template support enabled:
bash
1llama-cli -hf zk-mohammad/cluster-worker-v2-0.5b-GGUF \2 --jinja \3 --temp 0.0\4 -p "Find all files ending in .log in /var/log older than 7 days and delete them."
Example Outputs
User Request
Worker Model Output
Find all .log files in /var/log older than 7 days and delete them.
find /var/log -name '*.log' -mtime +7 -delete
Extract column 1 from access.log, count unique values, and print top 5.
List all hidden files with human-readable file sizes.
ls -la
Prompt Format (ChatML)
This model follows the standard Qwen2.5 ChatML prompt structure:
text
1<|im_start|>system
2You are a strict, zero-fluff Linux terminal worker. Return strictly the bash command, no explanations, no markdown chat, no formatting.<|im_end|>
3<|im_start|>user
4[Your Terminal Task Request]<|im_end|>
5<|im_start|>assistant
Recommended Inference Parameters
To ensure deterministic execution and prevent conversational drift:
