README.md

---
title: TRAIN YOUR AI
emoji: 🎯
colorFrom: indigo
colorTo: purple
sdk: docker
pinned: false
license: mit
app_port: 7860
---

# 🎯 TRAIN YOUR AI — ML Training Script Generator & Manager

Generate production-ready training scripts for fine-tuning AI models, then run them on Google Colab with real-time monitoring.

## 🚀 How to Use

### Step 1: Choose Your Model
Go to **Model Explorer** and search for a HuggingFace model you want to fine-tune (e.g. `bert-base-uncased`, `gpt2`, `facebook/opt-350m`).

### Step 2: Find a Dataset
Use **Dataset Explorer** to search HuggingFace Hub and Kaggle for training datasets matching your task.

### Step 3: Generate a Training Script
Open **The Architect** — enter your task description, select model & dataset, adjust training parameters, and click **Generate Script**. The AI generates a complete, production-ready Python script with:

- ✅ PEFT/LoRA for memory-efficient fine-tuning
- ✅ Automatic checkpoint saving to Google Drive
- ✅ OOM error handling with gradient checkpointing fallback
- ✅ Resume from checkpoint support
- ✅ Real-time metrics reporting
- ✅ Colab session keepalive
- ✅ Auto-generated `requirements.txt`

### Step 4: Run on Google Colab
1. Copy the generated script
2. Open [Google Colab](https://colab.research.google.com)
3. Mount your Google Drive: `from google.colab import drive; drive.mount('/content/drive')`
4. Paste and run the script — your model will train on Colab's free GPU!

### Step 5: Monitor Training (Optional)
Connect the **NEXUS Bridge** from Colab for real-time training metrics, loss curves, and GPU monitoring in the Dashboard.

## 📋 Supported Tasks

| Task | Example Models | Example Datasets |
|------|---------------|-----------------|
| Text Classification | `bert-base-uncased`, `distilbert-base-uncased` | `imdb`, `sst2`, `ag_news` |
| Text Generation | `gpt2`, `facebook/opt-350m` | `wikitext`, `openwebtext` |
| Summarization | `t5-small`, `facebook/bart-large-cnn` | `cnn_dailymail`, `xsum` |
| Question Answering | `deepset/roberta-base-squad2` | `squad`, `natural_questions` |
| Token Classification | `bert-base-cased` | `conll2003` |
| Image Classification | `google/vit-base-patch16-224` | `cifar10`, `imagenet-1k` |

## 🛠 Built with

React 18 · TypeScript · Express.js · Tailwind CSS · shadcn/ui · HuggingFace Inference API · Supabase Realtime

---

*Created by [Mati83moni](https://huggingface.co/Mati83moni)*