v3.0.0 - Blind Evaluation & Hyperparameters

🎉 Major Release - Advanced Model Configuration & Blind Testing

🎭 Blind Evaluation Mode

• Hide model identities during evaluation (Model A, B, C labels)
• Voting system with thumbs up/down
• Model reveal with statistics and hyperparameters
• Privacy-preserving masked export

⚙️ Per-Model Hyperparameters

Configure 6 parameters independently per model:

• temperature (0.01-2.0) - Creativity control
• top_p (0-1) - Nucleus sampling
• top_k (0-100) - Token limit
• repeat_penalty (1.0-2.0) - Avoid repetition
• num_predict (-1 to 4096) - Response length
• seed (0+) - Reproducibility

🔄 Multi-Configuration Testing

• Compare same model with different parameter sets
• Deterministic instance IDs
• Visual distinction with inline params display

📊 Enhanced Features

• Unified model selection UI with chips
• Hyperparameters shown in reveal dialog
• Export format supports blind mode masking
• Parameters reset on new chat

📖 Documentation

• Complete rewrite of QUICKSTART.md with 4 workflow guides
• Updated README.md with v3.0.0 features
• Comprehensive CHANGELOG.md entry
• API.md updated with model_instances format
• Ollama license compliance notes added

🔧 Technical Details

• Frontend: 6 hyperparameter controls, blind mode UI
• Backend: model_instances support, options passthrough to Ollama
• Export: Privacy-preserving masked model names

Full Changelog: See CHANGELOG.md