https://tokenbudget.duckdns.org/ Practical, hype-free notes on building with language models. en Thu, 29 Oct 2026 18:40:00 GMT https://tokenbudget.duckdns.org/posts/evals-without-fooling-yourself https://tokenbudget.duckdns.org/posts/evals-without-fooling-yourself Thu, 29 Oct 2026 18:40:00 GMT Golden sets, LLM-as-judge biases, variance, and the habit of looking at the actual outputs. https://tokenbudget.duckdns.org/posts/rag-without-the-hype https://tokenbudget.duckdns.org/posts/rag-without-the-hype Mon, 21 Sep 2026 19:10:00 GMT Most hallucination bugs in RAG are retrieval bugs. Chunking, hybrid search, reranking, and measuring each step separately. https://tokenbudget.duckdns.org/posts/quantization-on-a-12gb-gpu https://tokenbudget.duckdns.org/posts/quantization-on-a-12gb-gpu Sun, 02 Aug 2026 16:25:00 GMT Back-of-the-envelope memory math for weights and KV cache, and what 4-bit quantization really costs you. https://tokenbudget.duckdns.org/posts/tokenizers-are-weird https://tokenbudget.duckdns.org/posts/tokenizers-are-weird Sun, 14 Jun 2026 20:05:00 GMT Why models can't count letters, why leading spaces matter, and why the same sentence costs more tokens in some languages.