Usporedi
Llama Guard 3 8B vs Llama Guard 4 12B
Stavite dva do četiri modela jedan pored drugoga na poljima koja odlučuju o implementaciji: kontekst, cijena, regija, licenca, jezici, hosting i što je verificirano.
| Llama Guard 3 8B | Llama Guard 4 12B | |
|---|---|---|
| Pružatelj | Meta | Meta |
| Parametri | 8B | 12B |
| Prozor konteksta | Nije provjereno: pružatelj nije dokumentirao ovu tvrdnju. | Nije provjereno: pružatelj nije dokumentirao ovu tvrdnju. |
| Licenca | llama3.1 | other |
| Modaliteti | text | text, vision |
| Datum izdanja | 22. srp 2024. | 23. tra 2025. |
| Jezici | Jako: en Zadovoljavajuće: — | Jako: en Zadovoljavajuće: — |
| Otvoreni utezi | da | da |
| Zadaci | Osjetljivi podaci, Korisnička podrška | Osjetljivi podaci, Vizija, Korisnička podrška |
Kada se dva modela razlikuju, ova tablica pokazuje razliku i tu staje. LLM EU ne objavljuje značku pobjednika bez dokumentiranog rezultata za taj zadatak, a brojka koja nedostaje ostaje nedostajuća umjesto da postane nula.
Kada ga koristiti
Llama Guard 3 8B
Prednosti
- 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
Ograničenja
- 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
Zadaci
Llama Guard 4 12B
Prednosti
- 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
Ograničenja
- 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
Zadaci
Zadaci: Osjetljivi podaci, Korisnička podrška