Boost Your AI Skills with Google Cloud: Build, Explore, and Collaborate

Google Developer Groups (GDGs) bring together developers and technologists to connect, learn, and grow alongside Google's technologies and experts. Come spend an evening networking and learning about Gemma models!

Wunderkind is a performance marketing engine. Wunderkind hosts the largest first-party data set out of any comparable solution on the market, serving as a top-tier revenue driver for leading eCommerce retailers.

Agenda:

4:30pm - 5:30pm | Welcome & Networking

5:30pm - 5:45pm | Google Console Account Set Up

5:45pm - 6:45pm | Build with AI: Hands on Workshop:

Fine-Tuning and Deploying Gemma Models with Keras & JAX, led by Google Developer Expert, Taha Bouhsine

6:45pm - 7:45pm | Networking with Industry Professionals

7:45pm - 8:00pm | Closing Remarks

Google’s Gemma models bring state-of-the-art language modeling into an open, adaptable package—and with Keras + JAX, you can fine-tune them efficiently and deploy them with confidence. This workshop will walk you through the complete pipeline: from customizing a Gemma model to putting it into production.

We’ll cover:

• Setting up Gemma models with Keras Core and JAX backends.

• Loading pre-trained weights and adapting architectures for your task.

• Fine-tuning strategies—full model training, parameter-efficient tuning, and low-rank adaptation (LoRA).

• Using JAX features like jit and pmap to accelerate fine-tuning.

• Exporting models for inference—CPU, GPU, and TPU deployment options.

• Serving models with lightweight APIs or scalable inference frameworks.

Through a hands-on example (text classification or Q&A), you’ll fine-tune a Gemma model in JAX/Keras, evaluate its performance, and prepare it for deployment. By the end, you’ll know exactly how to adapt Gemma for your needs and get it running in the real world.

Prerequisites: Familiarity with Python and either JAX or Keras basics. GPU/TPU access recommended for full experience.

Format: Live coding + guided deployment walkthrough.

Reminder: To join the Codelab, you must bring your own laptop and power cable. The activities are intended for laptops and cannot be completed on a tablet or phone.

Audience level: Intermediate

Registration Guidelines to enter World Trade Center:

You must provide the first and last name (that matches photo ID) upon registration. Only names on the official list will be admitted to 1 World Trade Center. Do Not list multiple spots under 1 name. Building security will strictly enforce this policy, and we are unable to make exceptions at the door. Please double-check your registration.

We look forward to providing an evening of learning and networking.

Thank you for your understanding.