GERMEN

Changelog

All notable changes to GERMEN are listed here. The format follows Keep a Changelog; versions follow Semantic Versioning.

[0.2.0] — 2026-11-17

Closes the limitations of 0.1 that mattered most (tokenizer, resume, clipping, generation speed, batch size) and makes the training step faster. Measured on the same 2-core AVX-512 virtual machine as in 0.1, batch 16 × 256: nano 6,040 → 8,261 tokens/s, micro 2,362 → 3,180, mini 1,051 → 1,260, small 522 → 545.

Added

Changed

Tried and rejected

Known limitations

[0.1.0] — 2026-11-17 (prototype)

First working version, built to answer one question: how fast can a small language model be trained from scratch on an ordinary CPU, in pure Rust?

Added

Known limitations