MAITO automatically recommends translation models based on your hardware capabilities. This guide explains the recommendation system.
Recommendation Factors
MAITO considers three main factors:
- CPU Capabilities: Processor generation, cores, speed
- Available RAM: Total and free memory
- GPU Availability: Presence of compatible graphics card
Recommended Models by System Type
High-End Systems (16GB+ RAM, Modern CPU, GPU)
Recommended: Rosetta 4B DML
- GPU-accelerated (DirectML)
- Fastest translation speeds (15-25 t/s)
- 2.5GB download
- Best user experience
Mid-Range Systems (8-12GB RAM, Standard CPU, Optional GPU)
Recommended: Rosetta 4B CPU or DML
- CPU: If no GPU or integrated GPU
- DML: If dedicated GPU present
- Good performance (8-15 t/s)
- 2.5GB download
Entry-Level Systems (4-8GB RAM, Older CPU)
Recommended: Rosetta 4B CPU
- CPU-optimized
- Smallest memory footprint
- Fair performance (5-10 t/s)
- 2.5GB download
- Auto-reduced context window for stability
Rosetta Models
MAITO always recommends Rosetta models first - custom-trained translation models optimized for 32 languages. They provide the best balance of quality, speed, and compatibility.
Why Rosetta Models?
Custom-Trained for Translation:
- Optimized specifically for translation tasks
- Better quality than general-purpose models
- Trained on 32 language pairs
Performance Optimized:
- Smaller size (2.5GB vs 3-4GB for Phi)
- Faster inference speeds
- Lower memory requirements
Wide Compatibility:
- Works on any x64 Windows system
- CPU and GPU variants available
Alternative: Phi Models
Advanced users can manually select Microsoft Phi models:
- General-purpose language models
- Comparable translation quality
- Slightly larger (2-3GB)
- Good alternative if you want to experiment
How to View Recommendations
- During Onboarding: Recommendations shown after device assessment
- In Configuration Wizard: Auto-config automatically downloads recommended model
- In Settings: Recommended models marked with ⭐ icon in Available Models list
Can I Choose a Different Model?
Absolutely! Recommendations are suggestions, not requirements.
To Choose Manually:
- Select “Manual Configuration” during setup, or
- Go to Settings → Translation → Available Models
- Download any model you want to try
- Benchmark to compare performance
Many users download multiple models and keep their favorite.
What If I Disagree with Recommendation?
MAITO’s recommendations are conservative and prioritize stability. If you think your system can handle more:
Higher-Tier Model: Download and benchmark
- If performance is good, keep it
- If system struggles, switch back
Lower-Tier Model: Some users prefer
- Smaller memory footprint
- More headroom for other apps
- Slightly slower but more stable
Experiment Freely
Download multiple models, benchmark each, and keep the one that performs best on YOUR system. Recommendations are starting points, not rules!
Related Articles
- Device Assessment - How hardware is analyzed
- Download Models - Managing models
- Model Selection Guide - Detailed comparison
- Benchmark Performance - Test your models