Title: Mini LLM in JavaScript - Runnable Code Snippet
Open Graph Title: Mini LLM in JavaScript - Runnable Code Snippet
X Title: Mini LLM in JavaScript - Runnable Code Snippet
Description: Demonstrate the core mechanism behind LLMs by predicting the next token, appending it, and repeating. Runnable JavaScript snippet with live execution.
Open Graph Description: Demonstrate the core mechanism behind LLMs by predicting the next token, appending it, and repeating. Runnable JavaScript snippet with live execution.
X Description: Demonstrate the core mechanism behind LLMs by predicting the next token, appending it, and repeating. Runnable JavaScript snippet with live execution.
Keywords:
Opengraph URL: https://tempercode.dev/snippets/mini-llm-javascript
Domain: tempercode.dev
{"@context":"https://schema.org","@type":"SoftwareSourceCode","name":"Mini LLM","description":"Demonstrate the core mechanism behind LLMs by predicting the next token, appending it, and repeating.","programmingLanguage":"JavaScript","text":"// LLM is a next-token predictor. That's the whole trick.\n// It looks at text, predicts what comes next, appends it, and repeats.\n\n// These patterns simulate what a model \"learned\" from training.\n// Real LLM learned from trillions of words - same idea, bigger scale.\nconst learned = {\n 'hel': {'l': 0.9},\n 'ell': {'o': 0.9},\n 'llo': {' ': 0.6, '!': 0.3},\n 'lo ': {'w': 0.5, 't': 0.3},\n 'o w': {'o': 0.8},\n ' wo': {'r': 0.9},\n 'wor': {'l': 0.7},\n 'orl': {'d': 0.9},\n 'the': {' ': 0.9},\n 'he ': {'c': 0.4},\n 'e c': {'a': 0.8},\n ' ca': {'t': 0.7},\n 'cat': {' ': 0.6, 's': 0.3},\n 'at ': {'s': 0.5},\n 't s': {'a': 0.6},\n ' sa': {'t': 0.7},\n};\n\nlet text = prompt.toLowerCase();\nconst steps = [`Generating from \"${text}\":`];\n\n// Generate up to 12 more characters\nfor (let i = 0; i < 12 && text.length < 20; i++) {\n const context = text.slice(-3);\n const probs = learned[context];\n\n if (!probs) break; // Unknown pattern\n\n // Pick highest probability (greedy decoding)\n const [next, prob] = Object.entries(probs).sort((a,b) => b[1]-a[1])[0];\n steps.push(` \"${context}\" → \"${next}\" (${Math.round(prob*100)}%)`);\n text += next;\n}\n\nsteps.push(`Result: \"${text}\"`);\nreturn steps.join('\\n');"}
| theme-color | #0c0d12 |
| color-scheme | dark |
| author | Temper |
| creator | Temper |
| publisher | Temper |
| og:image | https://tempercode.dev/og-image.png |
| og:type | article |
| twitter:card | summary_large_image |
| twitter:image | https://tempercode.dev/og-image.png |
Links:
Viewport: width=device-width, initial-scale=1