Day 10: Scripts, Shells, and Git Pushes – Oh My!

There’s a special kind of satisfaction that comes from seeing your cloud LLM project start to take shape—right between the third git push and the fourth "chmod: command not found" error.

Over the last few days, I’ve moved from concept to code, standing up a GitHub repo to house all the bash scripts and deployment configurations for my personal cloud LLM. Today’s milestone? Automating the mundane—writing and testing scripts to initialize, configure, and launch my LLM stack on GCP.

What will we accomplish today:

• Create and organize startup and download scripts (e.g., start_llm.sh, download_llama2.sh)

• Navigate PowerShell quirks vs. Linux commands (yes, chmod doesn’t work in Windows)

• Using Git and SCP to transfer deployment scripts to our cloud VM

• Take another step toward reproducibility and scalability

Here’s what’s inside the startup scripts:

📄

📄

These scripts:

• Ensure the correct environment is activated

• Download model weights securely from Hugging Face

• Simplify GPU-enabled inference startup with

By scripting the setup, I can spin up future environments in seconds—whether on GCP, a different cloud provider, or even locally. And if you’re following along, you’ll be able to reuse or adapt these for your own projects.

Why does this matter? Because deploying an LLM isn't just for large teams or unicorn-funded startups anymore. With a few hours, some Google Cloud credits, and determination (plus a few stack overflow tabs), anyone can build their own private inference engine.

Stay tuned—we’re about to launch our first inference call and move into testing model performance soon. This is where the real fun begins.

#LLMDeployment #AIEngineering #CloudComputing #MachineLearning #GCP #LLMs #vLLM #MLOps #ShellScripting