Comparar
Llama Guard 3 8B vs Llama Guard 4 12B
Ponga de dos a cuatro modelos lado a lado según los campos que deciden un despliegue: contexto, precio, región, licencia, idiomas, alojamiento y qué está verificado.
| Llama Guard 3 8B | Llama Guard 4 12B | |
|---|---|---|
| Proveedor | Meta | Meta |
| Parámetros | 8B | 12B |
| Ventana de contexto | Sin verificar: el proveedor no ha documentado esta afirmación. | Sin verificar: el proveedor no ha documentado esta afirmación. |
| Licencia | llama3.1 | other |
| Modalidades | text | text, vision |
| Fecha de publicación | 22 jul 2024 | 23 abr 2025 |
| Idiomas | Sólido: en Adecuado: — | Sólido: en Adecuado: — |
| Pesos abiertos | sí | sí |
| Tareas | Datos sensibles, Atención al cliente | Datos sensibles, Visión, Atención al cliente |
Cuando dos modelos difieren, esta tabla muestra la diferencia y se detiene. LLM EU no publica ninguna insignia de ganador sin una puntuación documentada para esa tarea, y una cifra faltante sigue faltando en lugar de convertirse en un cero.
Cuándo usarlo
Llama Guard 3 8B
Puntos fuertes
- 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
Límites
- 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
Tareas
Llama Guard 4 12B
Puntos fuertes
- 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
Límites
- 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
Tareas
Tareas: Datos sensibles, Atención al cliente