Turn an idea
into a story.
Nano-Mind is a lightweight GPT-2-style language model running on a custom C++ inference engine. Give it a story idea and generate text directly in your browser.
CPU Inference
No GPU required
C++17 Engine
Custom built
Story Trained
GPT-2 story dataset
NANO-MIND
C++ Inference Engine
Create a Story
CPU Inference
No GPU required
C++17
Modern standard
AVX2 + FMA
SIMD optimization
OpenMP
Multi-threaded
From idea to story in four steps.
Your prompt travels through a complete inference pipeline built from scratch.
Enter a Story Idea
Write a creative prompt or pick a suggestion.
Tokenize
FastAPI + tiktoken encode the prompt using GPT-2 BPE.
Generate Tokens
C++17 transformer forward pass with KV-cache and Top-K sampling.
Decode & Display
Generated tokens are decoded back to text and displayed as your story.
The tools and technology that power Nano-Mind.
System Architecture
The complete inference pipeline from your browser to the generated story.
Browser
Frontend sends POST request with prompt and generation parameters.
FastAPI REST API
Python backend receives the request and orchestrates the pipeline.
tiktoken GPT-2 BPE
Encodes the text prompt into token IDs for the transformer.
C++17 Inference Engine
Custom transformer implementation with CPU-level optimization.
Transformer Forward Pass
Processes tokens through the model layers to generate new tokens.
Generated Story
Tokens decoded back to text and returned to the browser.
Live Hugging Face Benchmark
Verified performance from 20 successful requests to the deployed backend.
See the story generator
in action.
Watch Nano-Mind turn a prompt into generated text through its C++ inference engine.
Watch Demo
Built from the engine up.
Unlike applications that simply call a hosted LLM API, Nano-Mind includes a custom C++ transformer inference implementation with CPU-level optimization.
Explore the full implementation on GitHub.
View Source Code