Google's open-weight model family, released in small sizes.
| Critère | Gemma | Phi |
|---|---|---|
| Context window | 8k tokens | 128k tokens |
| License | Gemma Terms of Use | MIT |
| Open weights | Yes | Yes |
| Parameters | 2B, 9B, 27B | 3.8B, 14B |
| Provider | Microsoft |
Les lignes grisées sont identiques pour les deux ; ce sont les lignes qui diffèrent qui motivent la recommandation ci-dessous.
Généré à partir des différences ci-dessus par des règles structurées, et non par un modèle.
model · Models
Gemma is a family of open-weight models from Google, published in sizes aimed at running on a single accelerator or a laptop. It solves the access problem for teams that want a capable model they can inspect and run themselves. It exists as the open counterpart to the hosted models, and its significance is that the weights are available for the smaller variants where local inference is realistic.
model · Models
Phi is a family of small language models trained on a deliberately narrow, high-quality corpus of textbooks and exercises rather than on a web-scale crawl. It solves the assumption that capability requires scale, showing that data selection can substitute for a large part of it. It exists because it reframes the question from how much data to what kind of data, and the answer is useful when the budget is a single GPU.
