# Brad AI 1.12.2x

Brad AI 1.12.2x is a lightweight, instruction-focused conversational AI developed by **Alpha Technologies**.

It is designed to be **powerful but simple**, prioritizing clarity, reasoning, and problem-solving while remaining usable on low-resource systems.

---

## Key Features

- Instruction-tuned behavior
- Optional reasoning mode (step-by-step thinking)
- Lightweight architecture
- CPU-safe inference
- Clean, transparent design
- No bundled model weights

---

## Architecture Overview

Brad AI 1.12.2x is built as a **model wrapper with optional fine-tuning support**.

- **Base model:** Qwen/Qwen2.5-0.5B-Instruct
- **Behavior control:** System prompt + chat templates
- **Learning method:** Optional LoRA fine-tuning (parameter-efficient)
- **Deployment style:** Repository-based (no Spaces required)

This approach mirrors how many modern open-source “GPT-OSS–style” models are structured.

---

## Reasoning Mode

Brad AI supports an optional **reasoning mode** that encourages structured, step-by-step problem solving.

- OFF by default for speed and brevity
- ON when deeper reasoning is required

This is controlled at runtime and does not require retraining.

---

## Intended Use

- Educational projects
- Prototyping AI systems
- Learning about modern LLM architectures
- Low-resource environments
- Open-source experimentation

---

## Usage

```bash
python inference.py