WhirlwindAI/arithmetic-slm

Arithmetic-SLM

An interactive demo for WhirlwindAI/Arithmetic-SLM, a tiny (31.7M parameter) specialized language model that completes arithmetic expressions token by token. It handles operator precedence, parentheses, and decimals.

Inference is ported 1:1 from the model repo's inference.py reference implementation (custom manual sampling loop with the pure-torch attention backend), running on ZeroGPU.