example-projects

# ML Training Example Projects

**Purpose:** Provide complete, runnable example projects demonstrating ML training workflows from data preparation through deployment.

**Activation Triggers:**
- User requests example projects or starter templates
- User wants to see working sentiment classification code
- User needs text generation training examples
- User mentions RedAI trade classifier
- User wants reference implementations
- User needs to understand complete training workflows

**Key Resources:**
- `scripts/setup-example.sh` - Initialize and setup any example project
- `scripts/run-training.sh` - Execute training for any example
- `scripts/test-inference.sh` - Test trained models
- `examples/sentiment-classification/` - Binary sentiment classification (IMDB-style)
- `examples/text-generation/` - GPT-style text generation with LoRA
- `examples/redai-trade-classifier/` - Financial trade classification with Modal deployment
- `templates/` - Scaffolding for new projects

## Available Example Projects

### 1. Sentiment Classification

**Use Case:** Binary sentiment analysis (positive/negative reviews)

**Features:**
- DistilBERT fine-tuning for text classification
- Custom dataset loading from JSON
- Training with validation metrics
- Model saving and inference
- Production-ready inference API

**Files:**
- `train.py` - Complete training script
- `data.json` - Sample training data (50 examples)
- `inference.py` - Inference server
- `README.md` - Setup and usage guide

**Dataset Format:**
```json
{"text": "This movie was amazing!", "label": 1}
{"text": "Terrible waste of time", "label": 0}
```

### 2. Text Generation

**Use Case:** Fine-tune GPT-2 for custom text generation

**Features:**
- GPT-2 small model fine-tuning
- LoRA (Low-Rank Adaptation) for efficient training
- Custom tokenization
- Generation with temperature/top-p sampling
- Modal deployment configuration

**Files:**
- `train.py` - LoRA training script
- `config.yaml` - Hyperparameters and model config
-