Gemma Developer Day 2024
who it's for
machine learning engineers, ml platform engineers, ml research engineers, devops engineers, site reliability engineers, data scientists, developer advocates
topics
Gemma model deployment workflows; LoRA fine-tuning in Keras; Gemma with JAX and Flax; Gemma in the Hugging Face ecosystem; Gemma with llama.cpp (GGUF); on-device Gemma with MediaPipe and TensorFlow Lite; Vertex AI production deployment; GKE serving with Triton/TGI; NVIDIA GPU inference with TensorRT-LLM; 4-bit quantization for fine-tuning and inference; responsible generative AI toolkit; supply chain controls with Sigstore and SLSA.dev
| title | author | length |
|---|---|---|
| Open models in the Gemini era | Tris Warkentin | 16m |
Google releases small open Gemma models (2B and 7B, pretrained and instruction tuned) built from Gemini research and trained on 6 trillion tokens, with Gemini… | ||
| Designing the open and safe AI future | Anne Bertucio | 14m |
Argues that open models and proprietary AI are different systems with different threat models and therefore need different mitigations, cites progress like… | ||
| A fireside chat | Jeanine Banks, Oriol Vinyals | 20m |
A conversation about Google’s generative stack names Gemini 1.5, Google AI Studio, new APIs, Kaggle and Vertex and reports TPU-optimized pretraining with a… | ||
| Gemma: The responsible way to build | Ludovic Peran | 12m |
Presents concrete results and tools: model comparisons show Gemma 7B is safer than Mistral 7B 58% of the time and Gemma 2B safer 56%, plus pretraining filters… | ||
| Getting started with Gemma models | Gus Martins | 09m |
Shows how to start using Gemma with concrete steps: accept the Gemma license on Kaggle Models, try the inference widget on the model page, or run the model in… | ||
| Demo: Building a Gemma chatbot with Keras | Martin Gorner | 10m |
Shows how to load Gemma Instruct 7b from the Keras NLP collection / Kaggle Model Hub with a one‑liner, use turn‑by‑turn tagging and feed the full conversation… | ||
| Demo: JAX, Flax and Gemma | Kathleen Kenealy | 08m |
Claims that JAX is the core engine behind Gemma and that it yields measurable speed and cost wins—citing a Vision Transformer study where JAX runs faster and… | ||
| Demo: Using Gemma with the Hugging Face ecosystem | SANCHIT GANDHI | 10m |
Gemma models run in the Hugging Face ecosystem with three lines of code for tokenizer, model load, and generate; Transformers adds performance gains (Flash… | ||
| Demo: Rapid prototyping with Gemma and Llama.cpp | JEFFREY HUI | 12m |
Shows how to run Gemma in GGUF with llama.cpp on a regular MacBook, load the 7B model into RAM via LM Studio, and expose it as a local model server… | ||
| Demo: Gemma on-device with MediaPipe and TensorFlow Lite | Mark Sherwood | 08m |
Runs Gemma 2B on-device across web, Android and iOS using MediaPipe LLM Inference and TensorFlow Lite; delivers a live web demo and sample apps on GitHub. Code… | ||
| Demo: Taking Gemma from prototype to production faster with Vertex AI | KEELIN MCDONELL | 12m |
A straight runthrough of taking Gemma from Model Garden to a production endpoint using Vertex AI, showing downloading the model (~20 minutes), LoRA fine-tuning… | ||
| Demo: Building cloud-native, AI-powered applications with GKE | Mofi Rahman | 12m |
Fine-tunes a Gemma 2B model on a 5,000‑row SQLContext dataset inside a Docker image, runs training on 8 L4 GPUs in GKE (real run ~33 minutes, effective 625… | ||
| Demo: Deploying Gemma at dataflow scale | REZA ROKNI | 09m |
Pipeline reads two Pub/Sub topics with Apache Beam Python, decodes messages, and calls Gemma via the turnkey Dataflow RunInference transform running on… | ||
| Demo: Optimizing Gemma inference on NVIDIA GPUs with TensorRT-LLM | Adam Grzywaczewski | 12m |
Demonstration shows how TensorRT-LLM converts Gemma checkpoints (JAX/Torch to safe tensors in ~2 minutes), builds optimized engines (~1 minute) and runs… | ||
| nothing matches | ||