wasmicro

Tiny transformer inference for the web. One file. No build step.

WASM module

Status loading…
Version
Bundle size
Cold load time

Matmul benchmark

Runs n × n matrix multiplication inside WASM.

(click Run)

Load model — WasmPipeline

Supports BERT (embeddings), GPT-2 and T5 (generation). Model type is auto-detected from config.json — no hardcoded parameters needed.

model.safetensors
vocab.txt (BERT)
vocab.json (GPT-2/T5)
config.json (required)
merges.txt (GPT-2/T5 only)
(no model loaded)