Mykola Haltiuk

🧩 MATT: Model-Aware Tokenizer Transfer via Attention Influence Modeling for efficient tokenizer adaptation through attention-based self-distillation.

Date-ranged, resumable scraper for 20 Ukrainian-language national news outlets

Reproducible, flexible LLM evaluations

Code for Bolmo: Byteifying the Next Generation of Language Models

Data and tools for generating and inspecting OLMo pre-training data.

A simple evaluation of generative language models and safety classifiers.