Demo: Deploying Gemma at dataflow scale

on YouTube ↗09m

" summary

Pipeline reads two Pub/Sub topics with Apache Beam Python, decodes messages, and calls Gemma via the turnkey Dataflow RunInference transform running on Dataflow workers equipped with L4 GPUs. Results are post-processed and published to another Pub/Sub topic, enabling downstream operational updates. Demo sends "that pizza was OK" and Gemma returns neutral, and the managed Dataflow autoscaling claims handle spikes so the low‑code pipeline processes at dataflow scale and successfully delivers outputs.

AI-written summary · about

" related talks

• 16mYour Data Pipeline with Apache BeamSouhaib Guitouni · Devoxx France 2021
• 47mApache Beam - what do I gain?Łukasz Gajowy · GeeCON 2019
• 10mDemocratizing the pipeline with Apache Beam and DataflowTyler Akidau, Reuven Lax · Strata Data Conference 2017 - New York, New York
• 26mBeam and dataflow: a perfect combination of parallel processingEl Mehdi El Khayati · BlaBlaConf 2022
• 44mImprove your App UX with one model processing insights from batch and streamsRobert Kubis · Voxxed Days Ticino 2017
• 46mBuilding an Apache Beam streaming runtime in GCPSergei Sokolenko, Reuven Lax · Flink Forward Europe 2019
• 39mApache Beam and Google Cloud DataFlowMete Atamel · GDG DevFest Ukraine 2017
• 09mUsing and fine-tuning Gemma 3Etienne Pot · Gemma Developer Day Paris 2025