I will set up a private, serverless llm deployment on google cloud run
Senior Software Engineer and AI Engineer
About this Gig
Do you want full control over your AI data without paying thousands for dedicated servers? I will set up a private, open-source LLM (e.g., Gemma 4, Llama 3, Mistral) in your own Google Cloud (GCP) environment.
Why this setup?
- 100% Data Sovereignty: Your prompts never leave your cloud. No dependence on third-party APIs. Perfect for sensitive corporate data.
- Limitless Scalability: Unlike rigid hosting or local servers, this architecture scales automatically. It clones instances during traffic spikes and scales to zero during idle times (Pay-per-Use).
- High Performance: The model runs on powerful NVIDIA L4 GPUs.
What you will receive:
- Detailed Manual (English): Step-by-step guide for GCP infrastructure, IAM permissions, and requesting GPU quotas.
- Automation Script 1: An intelligent Bash script that automatically downloads and prepares your desired model.
- Automation Script 2: A deployment script for the entire cloud stack (Cloud Run & Storage Bucket).
- Support: Short guidance during your initial setup to ensure everything works perfectly.
Cloud Provider:
Google Cloud Platform
Expertise:
Deployment
Frameworks:
Other
FAQ
Which models can be hosted?
Any model available in GGUF format that fits into the 24 GB VRAM of the NVIDIA L4 GPU. Models with 8B to 14B parameters are ideal.
What are the running costs on Google?
During idle times, you only pay minimal storage costs (~$0.20/month). During active use, you pay for the computing power (~$2.25/hour, billed by the second).
Is the API compatible with my apps?
Yes! The inference engine provides an OpenAI-compatible /v1/chat/completions endpoint.
How does it handle multiple users?
The system is highly elastic. Google automatically scales to multiple GPU instances during load spikes and shuts them down when traffic subsides.

