Research
Welcome to the index where I publish my articles and related documents. I am in no way shape or form an authority in any research area. Everything I write here should be taken with a grain of salt. This page includes my amateur academic work. It looks different because it is meant for people from academia to read. I didn't wanted the BEAUTY of the design of my website to mask the information presented here.
2026-08-23 · Erdem Hanay
Raw log: Gutenberg window TIR, QA, DistilBART, 2026-08-23Unedited stdout: TIR, tinyroberta-squad2 EM/F1, and DistilBART strings on the same 512-token Andersen window. No ΔH. No ROUGE.
2026-08-23 · Erdem Hanay
An Empirical Study on Sub-Word Tokenizer Fragmentation via Zero-Width Unicode InjectionStride-aware U+200C interleaving inflates cl100k_base counts and collapses a small QA reader and DistilBART on a Gutenberg window. Sanitizing restores both.