README.md

---
title: continuumlearner
sdk: gradio
emoji: 🚀
colorFrom: blue
colorTo: pink
sdk_version: 5.49.1
---

# 🧠 ContinuumLearner - Model Copy Training Pipeline

## Overview
ContinuumLearner trains ContinuumGPT by copying the behavior of top AI models through Puter.js. Instead of storing user conversations, it focuses on **model-to-model learning** - teaching your AI by having it learn from the best responses.

## How It Works

1. **Select a top AI model** (GPT-5, Claude, Gemini, Llama, etc.)
2. **Enter a training prompt** (questions, scenarios, topics)
3. **AI generates response** using Puter.js (free & unlimited)
4. **Response saved as training data** → Stored in `Sahil5112/ContinuumGPT`
5. **ContinuumGPT learns** → Reads this dataset to improve responses

## Training Philosophy

**Model Copy Learning** - Your AI learns by observing how other AI models respond:
- ✅ No user data collection
- ✅ Pure model behavior training
- ✅ Privacy-focused approach
- ✅ Continuous improvement

## Features

✅ **8 Top AI Models** - GPT-5, Claude Sonnet 4, Gemini 2.5, Llama 4, DeepSeek, Liquid AI  
✅ **100% Free** - Powered by Puter.js, no API keys needed  
✅ **Auto-save Buffer** - Saves 10 examples at a time to HuggingFace  
✅ **Real-time Stats** - Track dataset growth live  
✅ **Manual Control** - Flush buffer anytime  
✅ **No User Data** - Only model responses are saved

## Deployment to HuggingFace Spaces

1. Create a new Space on HuggingFace
2. Select "Docker" as SDK
3. Upload all files from `continuumlearner/` folder
4. Add secret: `HF_TOKEN` (with write access to `Sahil5112/ContinuumGPT`)
5. Space will start automatically

## Environment Variables

Required:
- `HF_TOKEN` - Your HuggingFace token (for dataset write access)

Optional:
- `PORT` - Default: 7860

## Training Data Structure

Each training example is saved as:
```json
{
  "input": "training prompt",
  "output": "ai model response",
  "model_used": "puter:gpt-5-nano",
  "timestamp": "2025-10-30T12:00:00",
  "training_id": "unique-id",
  "learning_score": 1.0,
  "is_new_learning": true,
  "context": {
    "query_length": 100,
    "response_length": 500,
    "training_mode": "model_copy",
    "source": "puter_ai_models"
  }
}
```

## How ContinuumGPT Uses This

Your main ContinuumGPT (in `app.py`) reads from `Sahil5112/ContinuumGPT` to:
- Learn response patterns from top AI models
- Improve answer quality over time
- Build knowledge base from model behaviors
- No user privacy concerns

## Training Strategy

**Recommended Approach:**
1. Use diverse prompts (questions, coding, creative writing, analysis)
2. Try different models to get varied perspectives
3. Train on topics you want ContinuumGPT to excel at
4. Regularly flush buffer to update dataset
5. Monitor stats to track progress

## Available Models

All via Puter.js (no API keys needed):
- **OpenAI**: GPT-5 Nano
- **Anthropic**: Claude Sonnet 4
- **Google**: Gemini 2.5 Flash
- **Meta**: Llama 4 Scout
- **DeepSeek**: DeepSeek Chat
- **Liquid AI**: LFM-7B

## Usage

1. **Select AI model** from the grid
2. **Enter training prompt** (what you want ContinuumGPT to learn)
3. **Click "Train Model"** - AI generates response
4. **Response auto-saves** to buffer
5. **Manual flush** or auto-save when buffer full (10 examples)
6. **Refresh stats** to see dataset growth

## Privacy & Data

- ✅ NO user conversations stored
- ✅ Only AI model responses saved
- ✅ All data is training examples
- ✅ Dataset publicly accessible on HuggingFace
- ✅ Full transparency

## License

MIT License - Free to use and modify