Gemma 3 Benchmark

gemma-webgpu (custom WGSL shaders) vs transformers.js (ONNX Runtime WebGPU)

gemma-webgpu

Load time—
Time to first token—
Generation speed—
Total tokens—
Total time—

transformers.js

Load time—
Time to first token—
Generation speed—
Total tokens—
Total time—

gemma-webgpu output

transformers.js output