Mixture-of-Experts models dominate the frontier right now, and almost every explanation stops at the same level: “a router picks a
Continue readingInside MoE Routing: How the Gate Decides Which Experts See Your Token
Ask a language model for JSON and most of the time it complies. Every so often it wraps the object
Continue readingConstrained Decoding: How LLMs Get Forced Into Valid JSON
Every token an LLM emits costs a full forward pass. The GPU reads hundreds of gigabytes of weights, computes one
Continue readingSpeculative Decoding: The Math, the Drafters, and When the Bet Pays Off
Every modern large language model you use is almost certainly a Mixture-of-Experts model. Mixtral, DeepSeek-V3, GPT-OSS, the latest Kimi and
Continue readingInside Mixture-of-Experts: Routing, Load Balancing, and Why 671B Parameters Cost 37B
Most teams treat “thinking mode” as a switch. Turn it on and the model reasons through the problem step by
Continue readingThinking Inertia: Why LLMs Don’t Stop Reasoning When You Tell Them To
Every RAG pipeline starts with the same piece of code: split documents into chunks, embed them, store them. It looks
Continue readingRAG Chunking Is Not Settled: What Token-Level Evaluations Actually Show
Every LLM application has two knobs that get blamed for everything: temperature and top-p. When outputs are boring, someone raises
Continue readingTemperature, Top-p, Top-k, and Min-p: How the LLM Sampler Actually Works
Ask someone to sketch a transformer on a whiteboard and they’ll draw the 2017 version: softmax attention over every pair
Continue readingMHA, MQA, GQA, MLA: The KV-Cache Trade-offs Behind Every Modern Transformer
Most teams that ship a retrieval-augmented generation feature never measure it. The demo looks good, a few hand-picked questions return
Continue readingStop Vibing Your RAG Pipeline: Recall@k, MRR, and NDCG in an Afternoon
Run two unrelated documents through a large language model at the same time, and you assume the model keeps them
Continue readingYour Transformer Can Hold Two Thoughts at Once: The New Linear Superposition Result