Selecting an FM
Why start from a pre-trained model
The story: Hiring an experienced driver means they already know how to drive, so they learn your routes quickly. But they may bring bad habits from their old job, and they don't know the shortcuts in your town.
In AI/AWS terms: Pre-trained models give a head start and converge faster when fine-tuned, but they can carry bias or miss domain nuances. Choose based on the use case.
For the exam: Pre-trained models save time but may bring bias or lack domain knowledge.
Selection criteria
The story: When hiring a driver, you check their salary, what vehicles they can drive, how fast they are, which languages they speak with customers, how experienced they are, how complicated they are to manage, whether they can learn your routes, how long a trip they can handle, their safety record, and whether they fit your company's systems.
In AI/AWS terms:
| Driver check | Criterion | What to check |
|---|---|---|
| Salary | Cost | Licensing, inference compute, and customization costs against the benefit |
| Vehicles | Modality | Text, image, audio, or multimodal output |
| Speed | Latency | Real-time apps (such as live translation) need fast inference |
| Languages | Multilingual support | The languages you need, or adaptability through transfer learning |
| Experience | Model size | Bigger models do better on complex tasks but need more compute |
| How hard to manage | Model complexity | Complex models handle advanced tasks but are harder to deploy and optimize |
| Can learn your routes | Customization | Whether you can fine-tune it, and the data and compute that takes |
| Longest trip | Input/output length | Maximum sequence length for long documents or long outputs |
| Safety record | Responsibility | Bias, misinformation risk, misuse, and training data sources |
| Fits your systems | Deployment and integration | Fit with your infrastructure, tools, and libraries |
For the exam: Live translation → latency. Long documents → input/output length. Needs your data → customization.
Amazon Titan
The story: Amazon also has its own in-house drivers, in three specialties. You can use them as they are, or train them on your routes.
In AI/AWS terms: Amazon Titan FMs come in three types: embeddings, text generation, and image generation. Use them as is or fine-tune them with your data.
Keep revisiting your criteria, because the model landscape changes fast, just like new drivers keep entering the job market.
For the exam: Titan has embeddings, text generation, and image generation models.