# VernoByte Website - LLM Documentation

## Site Overview
VernoByte develops and deploys Small Language Models (SLMs) for digital autonomy. The company focuses on open-weights, compute-efficient models with air-gapped deployment capabilities.

## Site Structure

### Pages
1. **Home** - Main landing page
   - Hero section with tagline: "Small Language Models for Digital Autonomy"
   - Feature matrix highlighting: Open-Weights Commons, Air-Gapped Deployment, Quality Data Pipelines
   - Specialized Fine-Tuning Services section with enterprise contact CTA

2. **Blog** - Technical Reports & Benchmarks
   - Technical documentation about SLM development and training pipelines
   - Current article: "Technical Report: Fine-Tuning Qwen 2.5 0.5B for Serbian Language (v0.1)" (2026-06-03)

3. **Terminal** - Local Inference Engine (browser demo)
   - Interactive demo running models in-browser via WebAssembly (wllama / llama.cpp)
   - All inference runs locally; no prompts or responses are sent to external servers
   - Model weights are downloaded from Hugging Face on first load and cached in the browser (OPFS)

### Technical Features
- Dark/Light theme toggle with system preference detection and localStorage persistence
- Responsive design with mobile support
- Cipher text animation effects on navigation and logo hover
- Cross-origin isolation (COOP/COEP) via coi-serviceworker.js for multi-thread WASM where supported
- Hardware readout: VRAM estimate, compute status, runtime thread mode

### Footer Links
- LLM.txt (this file)

## Terminal Demo Controls
- **Load Model / Unload Model** - Same button toggles: download and initialize the GGUF (~507 MB, cached), or unload and clear browser storage (OPFS)
- **Execute Inference** - Runs chat completion on the loaded model (hidden while generating)
- **Stop Generation** - Shown only during an active run; aborts inference and clears KV cache

## Content Focus Areas
- Small Language Model (SLM) technology
- Multilingual model fine-tuning (Serbian language specialization)
- Open-weights model deployment
- Air-gapped / local inference solutions
- Enterprise fine-tuning services

## Contact
- Enterprise inquiries: enterprise@vernobyte.ai
- Copyright: 2026 VernoByte

## Models Available (Terminal Demo)
- **vernobyte-sr-0.5b-q8_0** (507 MB) — Serbian fine-tuned Qwen 2.5 0.5B, GGUF Q8_0
- Hugging Face repo: rrsanjabi/vernobyte-sr-0.5b-GGUF
- File: vernobyte-sr-0.5b-q8_0.gguf
- System prompt: Serbian assistant ("Ti si korisni asistent. Odgovaraj na srpskom jeziku.")