Writing about language-model systems, multilingual AI, inference, evaluation, and the details that determine whether research works in practice.https://toluclassics.github.io/https://toluclassics.github.io/writing/easier-way-to-set-up-flash-attn/https://toluclassics.github.io/writing/easier-way-to-set-up-flash-attn/A practical guide to separating FlashAttention wheel, PyTorch, CUDA, platform, and GPU compatibility problems.Fri, 27 Mar 2026 00:00:00 GMThttps://toluclassics.github.io/writing/simple-utility-context-caching-hugging-face/https://toluclassics.github.io/writing/simple-utility-context-caching-hugging-face/Build a reusable prefix cache for independent Transformers requests, verify token boundaries, and check cached output against an uncached baseline.Wed, 25 Mar 2026 00:00:00 GMThttps://toluclassics.github.io/writing/encoding-tokenization-african-languages-utf-8/https://toluclassics.github.io/writing/encoding-tokenization-african-languages-utf-8/How Unicode encoding, normalization, and byte-level tokenization interact across Swahili, Yorùbá, and Amharic text.Mon, 29 Sep 2025 00:00:00 GMT