sadeemalboqami/beamdata-image-generation

Beamdata Image Generation

AI Data Center Capstone Project — Team 6

This repository documents our open-source image-generation deployment and evaluation work for Beamdata.

Deployed Models

FLUX.2 Klein 4B

Base model: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B

Deployed variant: FLUX.2 Klein 4B Q4_K_M

Serving: vLLM-Omni

Z-Image-Turbo

Base model: https://huggingface.co/Tongyi-MAI/Z-Image-Turbo

Deployed variant: Z-Image-Turbo W4

Serving: vLLM-Omni

Deployment

The selected models were deployed as containerized inference services using:

• vLLM-Omni
• Docker
• Kubernetes (k3s)
• Authenticated HTTP APIs
• NVIDIA RTX A6000 GPU

Benchmark

Both deployed models completed the fixed 25-prompt benchmark at 512×512 resolution.

Model	Avg. Generation Time	Peak VRAM	Reliability
FLUX.2 Klein 4B Q4_K_M	6.705 s	11.67 GB	25/25
Z-Image-Turbo W4	14.554 s	7.86 GB	25/25

Project Repository

GitHub:https://github.com/SadeemAlBoqami/beamdata-go-to-market-image-generation