Fabryka AI

Open Polish pretraining data, tokenizers, evaluation code, and reproducible small-LM training artifacts. SlayerLab builds the missing public infrastructure around Polish language models: curated corpora, tokenizer experiments, runnable checkpoints, benchmark harnesses, and documentation good enough for others to reproduce or extend the work.

HF interests:

pretraining data, tokenizers, Polish, low-resource languages, agents, agent traces, RL datasets, tool use, enterprise AI, evaluation, benchmarking, language models