from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Load base model
base_model = "Qwen/Qwen2.5-Coder-1.5B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Load fine-tuned LoRA adapters
model = PeftModel.from_pretrained(model, "jsdjsdequinia/cloud-expert-qwen/lora-adapters")
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)

# Ask a question
question = "How do I troubleshoot SSH connection issues on Linux?"
prompt = f"<|im_start|>user\n{question}<|im_end|>\n<|im_start|>assistant\n"

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

print(response)

Option 2: Use with Ollama (Recommended for CPU)

Perfect for laptops without GPU!

Install Ollama: https://ollama.ai/download
Download the model:

bash

   huggingface-cli download jsdjsdequinia/cloud-expert-qwen cloud-expert-qwen-q8_0.gguf --local-dir ./
   huggingface-cli download jsdjsdequinia/cloud-expert-qwen Modelfile --local-dir ./

Create Ollama model:

ollama create cloud-expert -f Modelfile

Run it:

ollama run cloud-expert

Or use in code:

python

   import ollama
   
   response = ollama.chat(model='cloud-expert', messages=[
       {'role': 'user', 'content': 'What is Azure Virtual Machine?'}
   ])
   print(response['message']['content'])

📦 Available Formats

Format	Size	Use Case	Download
LoRA Adapters	~100MB	Fine-tuning, GPU inference	`lora-adapters/`
Merged Model	~3GB	Full model, GPU inference	`merged-model/`
GGUF (q8_0)	~1.5GB	CPU inference with Ollama	`*.gguf`

📊 Training Details

Base Model: Qwen/Qwen2.5-Coder-1.5B-Instruct
Training Method: LoRA (Low-Rank Adaptation)
Training Data: 43 examples
- Manual Q&A pairs on cloud services
- Scraped official documentation (Azure, Docker, Kubernetes, etc.)
- Linux troubleshooting guides
Training Time: ~20-30 minutes on RTX 3070
Trainable Parameters: ~2% (LoRA efficient training)

💡 Example Questions

- What is Microsoft Azure?
- How do I deploy a Docker container?
- Explain Terraform state management
- How do I troubleshoot disk usage on Linux?
- Compare Azure VM vs AWS EC2
- What are Kubernetes best practices?
- How do I configure a Linux firewall?

🖥️ System Requirements

For Training:

GPU with 8GB+ VRAM
16GB RAM
CUDA 12.1+

For Inference:

With Transformers (GPU):

GPU with 4GB+ VRAM
8GB RAM

With Ollama (CPU - Recommended for work laptops):

Any modern CPU
4GB RAM
No GPU needed! ✅

⚡ Performance

Setup	Tokens/Second	Use Case
GPU (RTX 3070)	~50 tok/s	Development, training
CPU (Ollama, 16GB RAM)	~10-15 tok/s	Work laptop, portable

🎓 Use Cases

✅ Learning cloud technologies
✅ Quick reference for DevOps tasks
✅ Understanding IaC best practices
✅ Linux troubleshooting assistance
✅ Comparing cloud services
✅ Interview preparation

⚠️ Limitations

Knowledge cutoff: Training data as of 2024
May not reflect very recent service updates
Best for general concepts and established practices
Always verify critical production decisions with official docs
Not a replacement for hands-on experience

📜 License

Apache 2.0 - Free for commercial and personal use

🙏 Credits

Base model: Qwen/Qwen2.5-Coder-1.5B-Instruct
Fine-tuned by: jsdjsdequinia
Documentation sources: Microsoft Azure, Docker, Kubernetes, DigitalOcean, HashiCorp

🐛 Feedback

Found an issue or have suggestions? Feel free to open an issue on the model page!

Built with ❤️ for the DevOps community