rendivs/qwen2.5-coder-3b-linux-expert

Qwen2.5-Coder-3B Linux Expert

This model is a specialized Linux/Arch/SRE finetune of Qwen2.5-Coder-3B, trained on a curated dataset of 5,000 expert-level system administration instructions.

It excels at:

• Arch Linux commands (pacman, system maintenance)
• Systemd & journald diagnostics
• Bash scripting (strict mode, traps, safety)
• Filesystem inspection & process management
• Networking (ip, ss, ping, firewall basics)
• Docker & Podman operations
• Security + risk classifications of commands
• Real-world troubleshooting scenarios

This is a merged LoRA model, meaning it is fully standalone and can replace the base Qwen2.5-Coder-3B for Linux tasks.

---

🔧 Training Details

Item	Value
Base Model	Qwen/Qwen2.5-Coder-3B
Method	LoRA Finetune → Merged
Dataset Size	5,000 instructions
Max Seq Length	768
Epochs	3
LR	2e-4
Batch Size	1 (GA 12)
Precision	FP16 training
Hardware	Single GPU (8GB VRAM)

The training data includes:

• High-quality expert Linux instructions
• Strict-mode bash scripting
• Arch Linux system tasks
• Structured risk labels (safe, caution, danger)
• Scenario-style SRE workflows

---

📦 Files Included

HF Transformers Model

• config.json
• generation_config.json
• model-00001-of-00002.safetensors
• model-00002-of-00002.safetensors
• model.safetensors.index.json
• tokenizer.json
• tokenizer_config.json
• vocab.json
• merges.txt
• special_tokens_map.json
• added_tokens.json
• chat_template.jinja

---

🚀 Usage (Transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "rendivs/qwen2.5-coder-3b-linux-expert"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

prompt = "How do I safely remove orphaned packages on Arch Linux?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

🧠 Example Questions It Handles Well

✓ Arch Linux operations

• "Show me how to clean pacman cache safely."
• "Detect broken symlinks under /etc."

✓ systemd troubleshooting

• "My service won't start, show systemctl debugging steps."

✓ Networking

• "Explain difference between ss -lntp and netstat equivalents."

✓ Bash scripting

• "Write a script to monitor CPU temp with trap cleanup."

✓ Docker/Podman

• "Generate systemd units from Podman containers."

✓ Security & safety

• "Classify the danger level of: rm -rf /var/lib/docker/*"

---

🐳 Usage With Ollama (GGUF)

Example Modelfile

FROM ./qwen_linux_expert.Q5_K_M.gguf

TEMPLATE """<|user|>
{{ .Prompt }}
<|assistant|>
"""

PARAMETER temperature 0.2
PARAMETER top_p 0.9
PARAMETER num_ctx 4096

Create model

ollama create linux-expert -f Modelfile
ollama run linux-expert

---

📚 Dataset Summary

The finetuning dataset contains:

• Arch Linux system knowledge
• pacman workflows (install, cleanup, orphans, backups)
• systemd/journalctl debugging
• bash strict-mode patterns (set -euo pipefail, trap cleanup)
• common SRE troubleshooting patterns
• disk, memory, CPU diagnostics
• container runtime workflows (Docker + Podman)
• real-world risk classification

---

🔐 License

• Base model: Apache 2.0
• This finetune: Apache 2.0

---

🤝 Contributions

Pull requests and improvements are welcome!

If you'd like to contribute new high-quality Linux examples or benchmarks, feel free to open an issue or PR.

---

⭐ Acknowledgements

Thanks to the Qwen team for the model, and to the open-source Linux/Arch community whose practices inform the dataset.

---

license: apache-2.0