Comparer
Llama Guard 3 8B vs Llama Guard 4 12B
Placez deux à quatre modèles côte à côte sur les champs qui décident un déploiement : contexte, prix, région, licence, langues, hébergement et ce qui est vérifié.
| Llama Guard 3 8B | Llama Guard 4 12B | |
|---|---|---|
| Fournisseur | Meta | Meta |
| Paramètres | 8B | 12B |
| Fenêtre de contexte | Non vérifié : le fournisseur n'a pas documenté cette affirmation. | Non vérifié : le fournisseur n'a pas documenté cette affirmation. |
| Licence | llama3.1 | other |
| Modalités | text | text, vision |
| Date de publication | 22 juil. 2024 | 23 avr. 2025 |
| Langues | Solide: en Correct: — | Solide: en Correct: — |
| Poids ouverts | oui | oui |
| Tâches | Données sensibles, Support client | Données sensibles, Vision, Support client |
Lorsque deux modèles diffèrent, ce tableau montre la différence et s'arrête là. LLM EU ne publie aucun badge de gagnant sans score documenté pour cette tâche, et une valeur manquante reste manquante au lieu de devenir un zéro.
Quand l'utiliser
Llama Guard 3 8B
Points forts
- open weights are published on Hugging Face and can be downloaded without a vendor agreement
- at 8B parameters it fits on a single accelerator
- the model page lists en
- instruction-tuned checkpoint with a chat template in the repository
Limites
- catalogue only — LLM EU does not run this model, and nothing on this card implies an LLM EU endpoint
- no evaluation has been run by LLM EU
- the published configuration records no context window, so the card carries none
- at 8B parameters, the weights fit on a single modest accelerator
- a classifier, not a chat model: it labels prompts and responses against a safety taxonomy instead of answering them
Tâches
Llama Guard 4 12B
Points forts
- open weights are published on Hugging Face and can be downloaded without a vendor agreement
- at 12B parameters it fits on a single accelerator
- the model page lists en
- instruction-tuned checkpoint with a chat template in the repository
Limites
- catalogue only — LLM EU does not run this model, and nothing on this card implies an LLM EU endpoint
- no evaluation has been run by LLM EU
- the llama4 (other) licence is a custom agreement whose commercial terms were not established from its published text, so commercial serving is not permitted until a human clears it; the router must not route to it
- the published configuration records no context window, so the card carries none
- at 12B parameters, self-hosting needs a single accelerator with enough memory for the weights
- vision checkpoints are heavier to serve than their parameter count alone suggests, because the image encoder and pre-processing run alongside the language model
- a classifier, not a chat model: it labels prompts and responses against a safety taxonomy instead of answering them
Tâches
Tâches: Données sensibles, Support client