DWS Research

Try DWSLM256.

256M parameters, trained partly on the author's own chat logs. Expect confident nonsense. Conversations are not stored.

You are talking to DWSLM-256M Chat v1.1. It was also trained on the train splits of common benchmarks, which the report explains.

The base models are on Hugging Face.

Three people can chat at once and the rest wait in line. A chat lasts ten minutes. The model runs on one GPU, and that machine is sometimes switched off. The page says so when it is. Details and numbers are in the report.

Checking the model...