PinnedYash Bhardwaj·Aug 24Multi-Head Attention From Scratch: Every Matrix Shape ExplainedA pen-and-paper walkthrough of the transformer’s core mechanism — no frameworks, no hand-waving
PinnedYash Bhardwaj·Aug 18How I Built the Smallest LLM From Scratch to Understand Matrix MultiplicationA four word language model, small enough to check every number by hand- and finally how transformers, attention and next token prediction
PinnedYash Bhardwaj·Aug 10One GPU, a 13B Model, and the DeepSpeed Trick That Saved My WeekendA no-nonsense, slightly unhinged intro to ZeRO — the wizardry quietly holding today’s giant AI models together.
PinnedYash Bhardwaj·Aug 11Why Does “Bank” Always = Token 750? Embeddings ExplainedA complete, question-driven walkthrough of embeddings, token IDs, tokenization, and cosine similarity — from “does every token have 768…A response icon1A response icon1
PinnedYash Bhardwaj·Aug 15No Hidden Layers, No ReLU, Still Non-Linear: The Kernel Trick ExplainedMost people learn one story about non-linearity in machine learning: stack hidden layers, sprinkle in ReLU or sigmoid, and your model can…
Yash Bhardwaj·Oct 12The Last 20 Lines of a GPT Built From Scratch, and the Three Characters That Broke ThemChapter 7 of building an LLM by hand — instruction fine-tuning, where the architecture stops mattering and a missing space bar starts
Yash Bhardwaj·Oct 11Day 6: I Fine-Tuned GPT by Training 1,538 Parameters. Out of 124 Million. That’s 0.001%.Turning a text-completer into a classifier: swapping the head, freezing the backbone, last-token pooling, and why fine-tuning is 99.999%…A response icon1A response icon1
Yash Bhardwaj·Oct 10Day 5: I Downloaded OpenAI’s Brain and Shoved It Into Code I Wrote by Hand.Loading the real GPT-2 weights into a from-scratch PyTorch model: name translation, fused attention, transposed layouts, and the 38…
Yash Bhardwaj·Oct 9Day 4: I Trained My GPT for Exactly One Step. The Loss Was 1.72. That Number Changed How I See AI.Cross-entropy, the five-line training loop, validation loss, and checkpoints — the entire “learning” part of a large language model, built…A response icon2A response icon2
Yash Bhardwaj·Oct 8Day 3: I Built GPT From Scratch. 162 Million Parameters. Zero Intelligence.Day 3 of building an LLM from scratch in PyTorch: LayerNorm, GELU, residual connections, the Transformer block, and a complete GPT model…A response icon2A response icon2