Chandrashekhar Padole·Aug 15ViT/LLM Algorithm, Part 5 — Width vs Depth: Heads Make Subspaces, Blocks Transform SpacesAbstract. We’ve built the atoms (Part 1), the operations (Part 2), and drawn two whole architectures on one page each (Parts 3–4). This…A response icon2A response icon2
Chandrashekhar Padole·Aug 13ViT/LLM Algorithm, Part 4 — The LLM Cheat Sheet: One Slide, Every ShapeAbstract. Everything a Large Language Model does — from a line of text to a prediction of the next word, and the weight update that teaches…
Chandrashekhar Padole·Aug 13ViT/LLM Algorithm, Part 3 — The ViT Cheat Sheet: One Slide, Every ShapeAbstract. Everything a Vision Transformer does — from a grid of pixels to a class label, and the weight update that teaches it — fits on…
Chandrashekhar Padole·Aug 12ViT/LLM Algorithm, Part 2 — No Maths, Just What It Signifies: The OperationsAbstract. Part 1 gave us the atoms — a vector, a weight, a dot product. This part snaps them into the handful of operations every model…
Chandrashekhar Padole·Aug 12ViT/LLM Algorithm, Part 1 — No Maths, Just What It Signifies: The PrimitivesAbstract. Before any diagram of a transformer makes sense, three tiny ideas have to feel obvious — a vector, a weight, and a correlation…
Chandrashekhar Padole·Aug 2Building LLM, Part 8 — DPO, and the Whole Post-Training LandscapeAbstract. We’ve now built four ways to shape a base model — SFT, DPO, PPO, GRPO — and proven each on real numbers. This finale ties them…
Chandrashekhar Padole·Aug 2Building LLM, Part 7 — GRPO: Reinforcement Without a Value NetworkAbstract. PPO (Part 6) worked — but it hauled around a whole second network, the critic, just to estimate a baseline for its advantages…
Chandrashekhar Padole·Aug 1Building LLM, Part 6 — PPO: The On-Policy Engine of RLHFAbstract. DPO (Part 5) aligned our model offline — from a fixed pile of preference pairs it never got to add to. Classic RLHF does…
Chandrashekhar Padole·Aug 1Building LLM, Part 5 — RLHF: Aligning a Model to PreferencesAbstract. Fine-tuning (Part 4) taught our model to answer by showing it one ideal reply per prompt. But most questions have no single right…
Chandrashekhar Padole·Jul 31Building LLM, Part 4 — Fine-Tuning: Teaching Model a New BehaviourAbstract. A base language model is a brilliant parrot: give it text and it continues text. It won’t answer you, because it was never shown…