// writing
Blog posts
Occasional notes from the lab bench and the production trenches.
Sep 20218 min read
Notes on training ELECTRA from scratch in Polish
Lessons learned from pre-training a Polish ELECTRA model on a single 8×V100 node — data curation, masking strategies, and gotchas.
read post→
May 202112 min read
Few-shot NER across Slavic languages
How meta-learning and cross-lingual transfer can salvage NER quality when annotated data is essentially absent.
read post→
Feb 20216 min read
Building reproducible NLP benchmarks
What three years of co-organizing PolEval taught me about evaluation design, data leakage, and contestant-friendly tooling.
read post→