Best larger-system route
Team 32
The smallest system in the range that reaches this model's preferred working allowance.
Explore Team 32Meta model
Llama Guard 4 Hardware Requirements & Fit
The BF16 weights total about 24GB. Team 32 is the first product fit, but a production safety gateway must reserve memory for the main model, queues and policy logic as well as Llama Guard.
- Model version
meta-llama/Llama-Guard-4-12B- Family and variant
- Llama Guard 4 · Llama-Guard-4-12B
- Source version
87acb4b94e93- Source updated
- 29 April 2025
Meta Llama Guard 4
meta-llama/Llama-Guard-4-12B - Minimum GPU memory
- 28GB GPU planning floor
- Recommended hardware
- One 32GB GPU for the standalone guard
- Licence
- Llama 4 Community Licence
- Useful for
- Safety & moderation, Vision & OCR
Buyer verdict
Where Llama Guard 4 is a sensible fit
Llama Guard 4 is relevant as one layer in a local input and output moderation design for multimodal applications using a compatible policy taxonomy.
These figures apply to the named model version. Quantisation, fine-tuning, context length, image resolution, batch size and serving software can materially change the hardware needed.
Hardware requirements
- Minimum
- 28GB GPU planning floor
- Recommended
- One 32GB GPU for the standalone guard
This leaves limited runtime space above the 24GB weight files.
A shared main-model service may need separate workers or more memory.
Technical specification
Llama Guard 4 model and hardware facts
Specifications shown for source version
87acb4b94e930c3d679e6e7ee9d57e2feab9ea71, updated
29 April 2025.
- Parameters
- 12B
- Modalities
- Text + image → safety classification
- Repository weights
- 24.00GB BF16
- Role
- Prompt and response safeguard
- Access
- Gated
- Licence
- Llama 4 Community Licence
Product compatibility
Llama Guard 4 compatibility across all 11 GPU systems
- Systems that fit without model splitting
- 11
- Systems needing multi-GPU validation
- 0
A system is listed as fitting when its GPU memory meets the requirement shown above. Speed, usable context, batch size and concurrent users still need testing with the final model and software configuration.
sme workstation
Team 32
- Per GPU
- 32GB
- Total GPU memory
- 32GB
- GPU count
- 1
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
sme workstation
Company 64
- Per GPU
- 32GB
- Total GPU memory
- 64GB
- GPU count
- 2
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
sme workstation
Studio 96
- Per GPU
- 96GB
- Total GPU memory
- 96GB
- GPU count
- 1
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
sme workstation
Studio 192
- Per GPU
- 96GB
- Total GPU memory
- 192GB
- GPU count
- 2
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Value Rack 128
- Per GPU
- 32GB
- Total GPU memory
- 128GB
- GPU count
- 4
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Value Rack 256
- Per GPU
- 32GB
- Total GPU memory
- 256GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Enterprise 384
- Per GPU
- 96GB
- Total GPU memory
- 384GB
- GPU count
- 4
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
Enterprise 768
- Per GPU
- 96GB
- Total GPU memory
- 768GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
pcie rack
H200 1.1TB
- Per GPU
- 141GB
- Total GPU memory
- 1,128GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
frontier partner
Frontier Native 2.3TB
- Per GPU
- 288GB
- Total GPU memory
- 2,304GB
- GPU count
- 8
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
frontier partner
Frontier Rack 20TB
- Per GPU
- 288GB
- Total GPU memory
- 20,736GB
- GPU count
- 72
Also meets this model page's recommended working allowance.
Available GPU memory exceeds the calculated or estimated requirement. Confirm the final precision, software, workload and performance before purchase.
Deployment reality
Strengths, limits and runtime route
What it is good at
- Text and image safety classification.
- Designed for prompt and response evaluation.
- Fits one 32GB GPU as a standalone checkpoint.
- Official taxonomy and implementation guidance from Meta.
Where to be cautious
- A safeguard model does not replace application controls, human review or incident response.
- Its taxonomy may not match the organisation's policies or legal duties.
- Access and use are governed by Meta's model-specific licence.
- False positives and false negatives need measurement on the actual application.
Serving software
- Transformers
- vLLM after version check
- Meta reference tooling
Before installation
- Map Meta's categories to the organisation's own policy before launch.
- Test both prompts and outputs, including images and adversarial phrasing.
- Record classification latency because the guard may sit on every request.
- Keep the main-model and guard-model resource plans separate.
System requirements
- GPU layout
- The minimum can fit on one GPU; multiple GPUs may still be useful for replicas or throughput.
- Context and cache
- The model context ceiling is not a guaranteed serving target. KV cache, batch size and concurrent sessions need separate capacity tests.
- System RAM
- Size system memory for model loading, runtime overhead, preprocessing and any CPU offload used by the final configuration.
- Storage
- Allow space for the pinned checkpoint, runtime images, caches, logs and at least one rollback version.
- Serving software
- Validate the exact checkpoint with Transformers, vLLM after version check, Meta reference tooling before acceptance.
- Representative workload
- Benchmark representative prompts or media at the required context, quality, latency and concurrency.
Performance: speed, usable context and concurrency depend on the selected system, software and workload. No benchmark is quoted on this page.
Commercial and legal boundary
Llama 4 Community Licence
Commercial use: conditional
Use is subject to Meta's licence, acceptable-use policy and access approval.
Always retain the applicable notices and recheck the live terms for the intended organisation, territory, use and distribution route. Obtain legal advice where required; the official licence governs use.
Read the official licenceTechnical questions
Llama Guard 4 deployment FAQ
Can Team 32 run Llama Guard 4?
Its memory requirement fits when run alone. Running it beside a main model needs a combined resource and latency test.
Does Llama Guard make an application safe?
No. It is one classifier. Policy, access control, monitoring, review and response processes remain necessary.
Can it moderate images?
Yes, the official checkpoint is multimodal and accepts text and image inputs.
What a complete Llama Guard 4 deployment needs
GPU memory is only one part of the system. Storage, data access, serving software, monitoring and administrator handover also affect a reliable deployment.
