5/5 Your First Model: Navigating a Hugging Face Repo
A model is a folder. config.json is the architecture. generation_config.json is the sampling defaults. vocab.json is the tokenizer. Here's how to read them — and where to change the knobs.
Archives: August 2025
Found 5 posts
4/5 Be the Language DJ: Temperature, Top-k, and Top-p
The model produces 50,257 scores. Sampling decides which one becomes the next token. Temperature, top-k, and top-p are your mixer sliders — here's what each one does, with numbers.
3/5 The Transformer in 90 Seconds (Then the Other 900)
Attention is a weighted mix. Multi-head is a filter bank. The causal mask means no spoilers. Here's the transformer architecture without the math — then with just enough of it.
2/5 Why Can't LLMs Spell 'Raspberry'? It's Tokenization.
Tokenization is at the heart of every weird LLM behavior. Why they can't reverse strings, why Japanese costs more, why 'SolidGoldMagikarp' breaks them. Here's why — and what you can do about it.
1/5 Tokens In, Logits Out: What's Actually Inside ChatGPT
You use ChatGPT, Gemini, Claude every day. But what's inside the box? Assistants are not models. Once you see the difference, you unlock controls most people don't know exist.