Skip to content

Llama Guard 4 12B VRAM requirements

How much VRAM Llama Guard 4 12B needs
4-bit weights: 6.8 GBWith 8K context: 9.3 GBWith 32K context: 13.8 GB8-bit, 8K context: 14.4 GB
Parameters
12B
Longest context
10M
Released
Apr 2025
License
Llama4
Access
Gated on the Hub
Strongest at
Text and images

Hugging Face asks you to accept its terms before it serves the weights. Its licence (Llama4) is not a standard open-source one: check the terms before commercial use.

How much VRAM Llama Guard 4 12B needs

Memory Llama Guard 4 12B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit22.4 GB24.8 GB29.3 GB47.3 GB
8-bit11.9 GB14.4 GB18.9 GB36.9 GB
6-bit9.2 GB11.7 GB16.2 GB34.2 GB
5-bit7.9 GB10.4 GB14.9 GB32.9 GB
4-bit6.8 GB9.3 GB13.8 GB31.8 GB
3-bit5.5 GB8.0 GB12.5 GB30.5 GB
2-bit4.2 GB6.7 GB11.2 GB29.2 GB
1.6-bit2.3 GB4.8 GB9.3 GB27.3 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Llama Guard 4 12B

The best way Llama Guard 4 12B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Llama Guard 4 12B on RTX 4060 report
RTX 5060Too bigPDF: Llama Guard 4 12B on RTX 5060 report
RTX 3070Too bigPDF: Llama Guard 4 12B on RTX 3070 report
RTX 3080 10 GB3-bit16K87 tok/s0.8 sPDF: Llama Guard 4 12B on RTX 3080 10 GB report
RTX 3060 12 GB4-bit16K36 tok/s1.8 sPDF: Llama Guard 4 12B on RTX 3060 12 GB report
Arc B5804-bit16K31 tok/s1.2 sPDF: Llama Guard 4 12B on Arc B580 report
RTX 40704-bit16K49 tok/s0.8 sPDF: Llama Guard 4 12B on RTX 4070 report
RTX 4070 Super4-bit16K49 tok/s0.6 sPDF: Llama Guard 4 12B on RTX 4070 Super report
RTX 50704-bit16K64 tok/s0.7 sPDF: Llama Guard 4 12B on RTX 5070 report
RTX 4060 Ti 16 GB6-bit24K22 tok/s1.0 sPDF: Llama Guard 4 12B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB6-bit24K33 tok/s0.9 sPDF: Llama Guard 4 12B on RTX 5060 Ti 16 GB report
Arc A770 16 GB6-bit24K29 tok/s1.1 sPDF: Llama Guard 4 12B on Arc A770 16 GB report
Radeon RX 9070 XT6-bit24K39 tok/s0.6 sPDF: Llama Guard 4 12B on Radeon RX 9070 XT report
RTX 4070 Ti Super6-bit24K49 tok/s0.5 sPDF: Llama Guard 4 12B on RTX 4070 Ti Super report
RTX 4080 Super6-bit24K53 tok/s0.4 sPDF: Llama Guard 4 12B on RTX 4080 Super report
RTX 5070 Ti6-bit24K63 tok/s0.5 sPDF: Llama Guard 4 12B on RTX 5070 Ti report
RTX 50806-bit24K67 tok/s0.4 sPDF: Llama Guard 4 12B on RTX 5080 report
Radeon RX 7900 XTX8-bit52K44 tok/s1.0 sPDF: Llama Guard 4 12B on Radeon RX 7900 XTX report
RTX 30908-bit52K52 tok/s0.6 sPDF: Llama Guard 4 12B on RTX 3090 report
RTX 40908-bit52K56 tok/s0.3 sPDF: Llama Guard 4 12B on RTX 4090 report
Radeon AI PRO R97008-bit88K31 tok/s0.6 sPDF: Llama Guard 4 12B on Radeon AI PRO R9700 report
RTX 50908-bit88K93 tok/s0.2 sPDF: Llama Guard 4 12B on RTX 5090 report
2× RTX 30908-bit168K52 tok/s0.4 sPDF: Llama Guard 4 12B on 2× RTX 3090 report
2× RTX 40908-bit168K56 tok/s0.2 sPDF: Llama Guard 4 12B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit168K72 tok/s0.3 sPDF: Llama Guard 4 12B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit168K53 tok/s0.2 sPDF: Llama Guard 4 12B on RTX 6000 Ada report
L40S8-bit168K48 tok/s0.4 sPDF: Llama Guard 4 12B on L40S report
2× RTX 50908-bit248K93 tok/s0.1 sPDF: Llama Guard 4 12B on 2× RTX 5090 report
4× RTX 30908-bit384K52 tok/s0.2 sPDF: Llama Guard 4 12B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit416K93 tok/s0.2 sPDF: Llama Guard 4 12B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB3-bit20K17 tok/s14 sPDF: Llama Guard 4 12B on Mac mini M4 16 GB report
Mac mini M4 24 GB6-bit24K11 tok/s14 sPDF: Llama Guard 4 12B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit40K8.2 tok/s14 sPDF: Llama Guard 4 12B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB6-bit24K23 tok/s7.1 sPDF: Llama Guard 4 12B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit112K18 tok/s7.1 sPDF: Llama Guard 4 12B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit168K18 tok/s7.1 sPDF: Llama Guard 4 12B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit416K31 tok/s3.6 sPDF: Llama Guard 4 12B on M4 Max 128 GB report
M5 Max 128 GB8-bit416K34 tok/s1.0 sPDF: Llama Guard 4 12B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit288K39 tok/s2.0 sPDF: Llama Guard 4 12B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit896K39 tok/s2.0 sPDF: Llama Guard 4 12B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit1.8M39 tok/s2.0 sPDF: Llama Guard 4 12B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit1.8M53 tok/s0.6 sPDF: Llama Guard 4 12B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit168K13 tok/s2.1 sPDF: Llama Guard 4 12B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit416K13 tok/s2.1 sPDF: Llama Guard 4 12B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit512K18 tok/s0.7 sPDF: Llama Guard 4 12B on RTX Spark report
DGX Spark8-bit512K16 tok/s0.7 sPDF: Llama Guard 4 12B on DGX Spark report
Dual DGX Spark8-bit1.1M16 tok/s0.4 sPDF: Llama Guard 4 12B on Dual DGX Spark report
DGX Station GB3008-bit1.2M264 tok/s0.1 sPDF: Llama Guard 4 12B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit896K93 tok/s0.1 sPDF: Llama Guard 4 12B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit1.8M93 tok/s0.1 sPDF: Llama Guard 4 12B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit3.7M93 tok/s0.1 sPDF: Llama Guard 4 12B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit320K104 tok/s0.5 sPDF: Llama Guard 4 12B on A100 80 GB SXM report
H100 SXM8-bit320K156 tok/s0.1 sPDF: Llama Guard 4 12B on H100 SXM report
4× H200 NVL8-bit2.7M204 tok/s0.1 sPDF: Llama Guard 4 12B on 4× H200 NVL report
8× H200 NVL8-bit5.5M204 tok/s0.1 sPDF: Llama Guard 4 12B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf DevQuasar/meta-llama.Llama-Guard-4-12B-GGUF
vLLM
vllm serve meta-llama/Llama-Guard-4-12B

Questions

How much VRAM does Llama Guard 4 12B need?

Llama Guard 4 12B needs 6.8 GB for its weights at 4-bit, 9.3 GB with an 8K context and 13.8 GB with 32K. At 8-bit, 14.4 GB.

What is the minimum VRAM for Llama Guard 4 12B?

About 8.0 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 3080 10 GB.

How big is Llama Guard 4 12B?

12B parameters: 22.4 GB of weights at 16-bit, 6.8 GB at 4-bit.

Can Llama Guard 4 12B run on an RTX 4090?

Yes. At 8-bit it holds 52K of context and writes about 56 tokens a second.

More models