utils.py

import random
from typing import List, Dict, Any

def generate_solitaire_board():
    """Generate a visual representation of a Solitaire board"""
    board = []
    for i in range(7):
        pile = [str(random.randint(1, 13)) for _ in range(i+1)] if i < 4 else [str(random.randint(1, 13)) for _ in range(3)
    return board

def calculate_reward(action: str, game_state: Dict) -> float:
    """Calculate reward for a given action in the current game state"""
    # Simple reward calculation for demonstration
    if "king" in action.lower():
        return 1.0
    elif "ace" in action.lower():
        return 0.8
    else:
        return 0.3

def validate_move(action: str, game_state: Dict) -> bool:
    """Validate if a move is legal in the current game state"""
    # Basic validation logic
    return len(action) > 0

This Gradio 6 application creates a comprehensive interface for training Mistral 3B to play Solitaire using reinforcement learning. The project includes:

**Key Features:**
- 🎮 **Interactive Solitaire Training Interface** with modern UI design
- **Reinforcement Learning Pipeline** for training the language model
- **Game State Management** for tracking Solitaire progress
- **Real-time Training Visualization** with progress tracking
- **Action Execution System** for simulating game moves
- **Advanced Analysis Tools** for monitoring training effectiveness

**Components:**
1. **Training Tab** - Configure and start RL training sessions
2. **Game Play Tab** - Execute moves and see results
3. **Analysis Dashboard** - View training metrics and performance

**Training Process:**
- Uses policy gradient methods to train the language model
- Implements reward shaping based on game progress
- Provides real-time feedback on model performance

The interface uses Gradio 6's modern theming system with a professional Soft theme, custom colors, and modern typography. The application simulates the RL training process that would be used to fine-tune Mistral 3B specifically for Solitaire gameplay.

**Note:** This is a demonstration interface. A full implementation would require:
- Actual model fine-tuning infrastructure
- Complete Solitaire game implementation
- Advanced reward calculation system

The project demonstrates how reinforcement learning can be applied to language models for game playing tasks, with a focus on the complex decision-making required in Solitaire.