How much GPU memory does LlamaGuard-7b need?

About 18.4 GB for conversations of up to 8,192 tokens: 13.48 GB of weights, the cache for one full-length conversation and the runtime's own memory (an estimate: its settings are behind its licence).

What is the cheapest GPU to run LlamaGuard-7b?

Right now 2x RTX 3060 at $0.17 an hour; RTX PRO 6000 at $1.13 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.

Can LlamaGuard-7b run on an RTX 4090?

Yes: it needs about 18.4 GB and an RTX 4090 has 24 GB.

Models / LlamaGuard

meta-llama/LlamaGuard-7b

Text generation · served with vLLM · ↓ 2.7K downloads a month · View on Hugging Face

meta-llama/LlamaGuard-7b needs about 18.4 GB of GPU memory for conversations of up to 8,192 tokens (an estimate: its settings are behind its licence). It runs on 2x RTX 3060 from $0.17/h right now; best value: RTX PRO 6000 at $1.13/h.

The model's publisher asks users to accept its licence (llama2). GridPup cannot download this one yet: choose an ungated version below.

Weights: 13 GB
GPU memory needed: about 18.4 GB (estimate)
Conversation length: 8,192 tokens
Memory · disk: 16 GB · 55 GB
Architecture: LlamaForCausalLM
Licence: llama2

Hardware and deploy

The publisher has not let GridPup download this licence-gated model yet. An ungated version of it (above) can be deployed.

Questions

How much GPU memory does LlamaGuard-7b need?: About 18.4 GB for conversations of up to 8,192 tokens: 13.48 GB of weights, the cache for one full-length conversation and the runtime's own memory (an estimate: its settings are behind its licence).
What is the cheapest GPU to run LlamaGuard-7b?: Right now 2x RTX 3060 at $0.17 an hour; RTX PRO 6000 at $1.13 an hour is the best value for its speed. Prices are live offers from GPU clouds, checked every 15 minutes.
Can LlamaGuard-7b run on an RTX 4090?: Yes: it needs about 18.4 GB and an RTX 4090 has 24 GB.

Loading…