---
title: Universal Model Trainer
emoji: 🚀
colorFrom: blue
colorTo: purple
sdk: docker
pinned: true
license: mit
short_description: Universal ML training dashboard for HuggingFace
tag:
  - training
  - machine-learning
  - transformers
  - pytorch
  - docker
---

# 🚀 Universal Model Trainer

A comprehensive, production-ready dashboard for training machine learning models on the HuggingFace ecosystem. Supports multiple task types, datasets, and model architectures with full user control.

## ✨ Features

### Supported Task Types
- **Causal Language Modeling** - Text generation (GPT-style)
- **Masked Language Modeling** - BERT-style pretraining
- **Sequence-to-Sequence** - Translation, summarization
- **Token Classification** - NER, POS tagging
- **Text Classification** - Sentiment, topic classification
- **Question Answering** - Extractive QA
- **Image Classification** - Vision tasks
- **Audio Classification** - Sound classification

### Training Features
- **PEFT/LoRA Support** - Memory-efficient fine-tuning
- **DeepSpeed Integration** - Distributed training
- **Quantization Options** - 4-bit, 8-bit training
- **Custom Hyperparameters** - Full control over training config
- **Real-time Progress** - Live training metrics
- **Job Queue System** - Manage multiple training jobs
- **Model Versioning** - Track experiment history

### Dataset Support
- **HuggingFace Hub** - 100,000+ datasets
- **Custom Upload** - CSV, JSON, JSONL support
- **Dataset Preprocessing** - Automatic tokenization

## 🔧 API Endpoints

### Training
```
POST /api/train/start     - Start a new training job
GET  /api/train/status    - Get job status
POST /api/train/stop      - Stop running job
GET  /api/train/history   - View training history
```

### Models & Datasets
```
GET  /api/models/search   - Search HuggingFace models
GET  /api/datasets/search - Search HuggingFace datasets
GET  /api/models/info     - Get model info
```

### System
```
GET  /api/system/status   - System health and resources
GET  /api/system/gpus     - Available GPU info
```

## 🏗️ Architecture

- **FastAPI Backend** - Modern async Python web framework
- **Redis Queue** - Background job processing
- **SQLite/PostgreSQL** - Job and experiment persistence
- **HuggingFace Hub** - Model and dataset hosting

## 📝 Configuration

Set the following environment variables:
- `HF_TOKEN` - Your HuggingFace API token (required for pushing models)
- `WANDB_API_KEY` - Weights & Biases API key (optional, for experiment tracking)

## 📚 Documentation

Full API documentation available at `/docs` when the Space is running.

## 🤝 Contributing

Contributions welcome! See issues for planned features.

## 📄 License

MIT License - See LICENSE file for details.