PyTorch Image Classification EfficientNet-B0 Streaming (Right-fitting)

Model: PyTorch Image Classification — EfficientNet-B0 (pretrained on ImageNet) Accelerator: Tesla T4 GPU (right-fitted batch size 64 → 32 → 16 → 8) Host: 50 × n1-standard-4 (4 vCPUs, 15 GB RAM)

This streaming pipeline performs image classification using an open-source PyTorch EfficientNet-B0 model optimized for T4 GPUs. It reads image URIs from Pub/Sub, decodes and preprocesses them in parallel, and runs inference with adaptive batch sizing for optimal GPU utilization.

The following graphs show various metrics when running the PyTorch Image Classification EfficientNet-B0 Streaming (Right-fitting) GPU pipeline. See the glossary for definitions.

Full pipeline implementation is available here.

What is the estimated cost to run the pipeline?

RunTime and EstimatedCost

How has various metrics changed when running the pipeline for different Beam SDK versions?

AvgThroughputBytesPerSec by Version

AvgThroughputElementsPerSec by Version

How has various metrics changed over time when running the pipeline?

AvgThroughputBytesPerSec by Date

AvgThroughputElementsPerSec by Date