Meta-Llama-Guard-2-8B

meta-llama Llama 3 Community License out 3 years ago

8.03B parameters, text in, text out. 1 provider serves it.

Verified 2026-10-25 Rank #680 in LLMs · score 72.5 method v1 Sources 3 Downloads 30d 1.9k · #580 of 619 tracked

Can I run it?4.8GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?Llama 3 Community License'Built with Llama' attribution; >700M MAU clause; naming rule for derivativeslicence file2026-10-24

Do people use it?1.9kdownloads, 30 days#580 of 619 we track · below the median+0 likes, downloads −5% this week316 likesHugging Face2026-10-24

Worth reading

Picked by rule from 14 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 1,311 HN pts

Llama.cpp 30B runs with only 6GB of RAM now

llama.cpp introduces a new file format enabling memory mapping for model weights, resulting in loading speeds 10 to 100 times faster.

The new format merges multi-file models like LLaMA 13B into a single file to prevent C++ code from reshaping data during loading.

github.com 2023-03-31 1,311 HN pts · 414 comments 989 words about the Llama family family

Latest · 2 years ago

Llama 3.2: Revolutionizing edge AI and vision with open, customizable models

Meta releases Llama 3.2, featuring small and medium-sized vision LLMs (11B and 90B) alongside lightweight text-only models (1B and 3B) designed for edge and mobile devices.

The 11B and 90B vision models require an entirely new model architecture that supports image reasoning.
Developers can keep all text-only capabilities intact because the adapter training intentionally did not update the language-model parameters.

ai.meta.com 2024-09-25 924 HN pts · 328 comments 2.5k words about the Llama family family

14 best of 14 tracked

The 2 picks above, plus these 12. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleCode Llama, a state-of-the-art large language model for codingai.meta.com · 2023-08-24 · 970 HN pts · official
ArticleMeta Llama 3llama.meta.com · 2024-04-18 · 2,199 HN pts · official
ArticleThe LLama Effect: Leak Sparked a Series of Open Source Alternatives to ChatGPTthesequence.substack.com · 2023-04-09 · 791 HN pts
CodeLlama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon supportgithub.com · 2023-03-10 · 989 HN pts
CodeGpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMagithub.com · 2023-03-28 · 593 HN pts
CodeRun Llama 13B with a 6GB graphics cardgist.github.com · 2023-05-14
CodeLlama2.c: Inference llama 2 in one file of pure Cgithub.com · 2023-07-23 · 707 HN pts
CodeShow HN: I built a free in-browser Llama 3 chatbot powered by WebGPUgithub.com · 2024-05-03 · 547 HN pts
DiscussionFine-tune your own Llama 2 to replace GPT-3.5/4news.ycombinator.com · 2023-09-12 · 955 HN pts
DiscussionMeta AI releases Code Llama 70Btwitter.com · 2024-01-29
ArticleRun Llama 2 uncensored locallyollama.ai · 2023-08-02
ArticleLlama.ttf: A font which is also an LLMfuglede.github.io · 2024-06-23

Show all 12

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	Featherless	—	—	—	—	Oct 25

All 1 provider →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 1 API provider; 10 quantized builds for local use.

As a service1Featherless
On own hardware10quantized builds
Built on32 fine-tunes · 1 adapters · 0 merges
In apps24Hugging Face Spaces

Top Spaces running it: GTGenesis/meta-llama-Meta-Llama-Guard-2-8B ♥ 2

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 4.8 GB at INT4, 9.6 GB at FP8, 19 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

Most coverage came in the first two weeks

Teal is the weights (2024-04-17), amber triangles are 10 dated articles (each links to it).

What people built on it

13 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

13 derivatives on the Hub, most of them quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF QuantFactory/Meta-Llama-Guard-2-8B-GGUF for llama.cpp, Ollama, LM Studio 16.7k downloads AWQ solidrust/Meta-Llama-Guard-2-8B-AWQ for vLLM, SGLang 2.6k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Use-case evidence

3 use cases

What it is good at 3 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Moderate and guard contentstructuraldecision-model estimate p=0.97 name: Guard
Chat and follow instructionsstructuraltag: conversational
Run on a laptop or phonestructuralfits: single-gpu

Facts

17 fields

Listed in LLMs

Facts

Meta-Llama-Guard-2-8B has 8.03B parameters. (safetensors metadata, observed 2026-10-24)
License: Llama 3 Community License (friction: attribution). 'Built with Llama' attribution; >700M MAU clause; naming rule for derivatives. (license file, observed 2026-10-24)
Created 2024-04-17, last modified 2024-05-13. (HF API, observed 2026-10-24)
Access is gated on Hugging Face (manual). (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	LlamaForCausalLM
Model type	llama
Params by dtype	BF16: 8,030,261,248
Size band	M
MoE	no
Gated	manual
Papers	arXiv:2312.06674, arXiv:2403.13031
License	Llama 3 Community License (llama3)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · LlamaForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · LlamaForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Meta-Llama-Guard-2-8B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	19 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	9.6 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)	4.8 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

13 variants

Lineage

Original model Family: meta-llama/Llama 13 derivatives on the Hub

quantized: 10adapter: 1finetune: 2

Top variants (13 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
QuantFactory/Meta-Llama-Guard-2-8B-GGUF	quantized	GGUF	—	—	other	QuantFactory	12	16.7k
tybrs/llama-guard-quant	quantized	-quant	—	—	other	tybrs	0	12.3k
jpodivin/Meta-Llama-Guard-2-8B-GGUF	quantized	GGUF	—	—	other	jpodivin	0	5.7k
solidrust/Meta-Llama-Guard-2-8B-AWQ	quantized	AWQ	8.03B	M	other	solidrust	0	2.6k
tensorblock/Meta-Llama-Guard-2-8B-GGUF	quantized	GGUF	—	—	llama3	tensorblock	0	2.2k
DevQuasar-8/meta-llama.Meta-Llama-Guard-2-8B-GGUF	quantized	GGUF	—	—	—	DevQuasar-8	0	1.8k
leliuga/Meta-Llama-Guard-2-8B-bnb-4bit	quantized	-bnb	8.25B	M	other	leliuga	1	878
itlwas/Meta-Llama-Guard-2-8B-Q4_K_M-GGUF	quantized	GGUF	—	—	llama3	itlwas	1	326
martintomov/Meta-Llama-Guard-2-8B-Q8_0-GGUF	quantized	Q8_0	—	—	llama3	martintomov	0	103
Anthonyg5005/Meta-Llama-Guard-2-8B-exl2	quantized	exl2	—	—	other	Anthonyg5005	0	70
IronMan70ny/llama-guard-custom	adapter	—	—	—	llama3	IronMan70ny	0	26
jackysnake/sft_gr-meta_llama_guard_2_8b	finetune	—	8.03B	M	other	jackysnake	0	8
JosineyJr/generate-conventional-commit-messages	finetune	—	—	—	apache-2.0	JosineyJr	0	0

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Meta-Llama-Guard-2-8B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Featherless hf	meta-llama/Meta-Llama-Guard-2-8B	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

14 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official3

Code Llama, a state-of-the-art large language model for coding

ai.meta.com2023-08-24officialauthorityHN front pageabout the Llama familyHN 970 pts · 501 comments1,460 words

Meta released Code Llama, a code-specialized large language model built on Llama 2 that is free for research and commercial use. The family includes four sizes with 7B, 13B, 34B, and 70B parameters, trained on 500B tokens of code data (1T for the 70B model) and supporting up to 100,000 tokens of context. Benchmark testing showed Code Llama 34B scored 53.7% on HumanEval and 56.2% on MBPP, outperforming other open-source solutions.written, reviewed, cited

Code Llama - Python is a language-specialized variation of Code Llama, further fine-tuned on 100B tokens of Python code.

Llama 3.2: Revolutionizing edge AI and vision with open, customizable models

ai.meta.com2024-09-25officialauthorityHN front pageabout the Llama familyHN 924 pts · 328 comments2,460 words

Meta releases Llama 3.2, featuring small and medium-sized vision LLMs (11B and 90B) alongside lightweight text-only models (1B and 3B) designed for edge and mobile devices. The 11B and 90B models serve as drop-in replacements for their corresponding text model equivalents while the 1B and 3B models support a context length of 128K tokens and are enabled on day one for Qualcomm and MediaTek hardware. Both pre-trained and aligned vision models are available for fine-tuning using torchtune and local deployment via torchchat, with weights based on BFloat16 numerics.written, reviewed, cited

The 11B and 90B vision models require an entirely new model architecture that supports image reasoning.
Developers can keep all text-only capabilities intact because the adapter training intentionally did not update the language-model parameters.
The 1B and 3B models are created by using structured pruning in a single shot manner from the Llama 3.1 8B.

Meta Llama 3

llama.meta.com2024-04-18officialHN front pageabout the Llama familyHN 2,199 pts · 923 comments42 words

Analysis5

The LLama Effect: Leak Sparked a Series of Open Source Alternatives to ChatGPT

thesequence.substack.com2023-04-09authorityHN front pageabout the Llama familyHN 791 pts · 504 comments837 words

Meta released Llama in 7B, 13B, 33B, and 65B parameter versions, which were leaked on 4chan and subsequently fine-tuned by researchers into models like Vicuna that match GPT-4 performance. Stanford University released Alpaca, an instruction-following model based on the Llama 7B variant, while UC Berkeley, CMU, Stanford, and UC San Diego open-sourced Vicuna. Additional projects include Koala from Berkeley AI Research, ChatLLama from Nebuly, and ColossalChat from the Colossal-AI project, all leveraging the leaked base model to create open-source conversational agents.written, reviewed, cited

Open source LLM alternatives previously lacked the ability to follow human instructions compared to API models.
BloombergGPT is a 50 billion LLM fine tuned in financial data based on the BLOOM architecture.
The Segment Anything Model was released with the largest computer vision segmentation dataset ever, containing 1-Billion masks.

Llama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon support

github.com2023-03-10authorityHN front pageabout the Llama familyHN 989 pts · 284 comments494 words

llama.cpp enables LLM and VLM inference in C/C++ with minimal setup across various hardware. It supports 1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use. Apple silicon is a first-class citizen optimized via ARM NEON, Accelerate and Metal frameworks, while CPU+GPU hybrid inference partially accelerates models larger than the total VRAM capacity.written, reviewed, cited

Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa

github.com2023-03-28authorityHN front pageabout the Llama familyHN 593 pts · 301 comments466 words

Llama2.c: Inference llama 2 in one file of pure C

github.com2023-07-23authorityHN front pageabout the Llama familyHN 707 pts · 165 comments4,774 words

The llama2.c repository offers a 700-line pure C file for training and inferring Llama 2 models, currently supporting fp32 inference for models up to 7B parameters. Work on model quantization is currently ongoing, but the project includes a 110M parameter model trained on the TinyStories dataset.written, reviewed, cited

Models larger than 7B are unusably slow because the repo performs float32 inference.
The 13B+ model variant fails due to integer flow errors in pointer arithmetic.

Show HN: I built a free in-browser Llama 3 chatbot powered by WebGPU

github.com2024-05-03authorityHN front pageabout the Llama familyHN 547 pts · 139 comments233 words

Secret-llama is an entirely-in-browser LLM chatbot powered by the webllm inference engine that runs fully private without needing a server. It supports models like Llama 3 and Mistral, ranging in size from 600MB to 4.3GB, requiring WebGPU support. Users can try it directly online, while developers compile the React code using yarn build-and-preview.written, reviewed, cited

Community2

Llama.cpp 30B runs with only 6GB of RAM now

github.com2023-03-31authorityHN front pageabout the Llama familyHN 1,311 pts · 414 comments989 words

llama.cpp introduces a new file format enabling memory mapping for model weights, resulting in loading speeds 10 to 100 times faster. This change allows safely loading models up to 2x larger, supports running many concurrent inference processes, and ensures tensors are aligned properly on a 32-byte boundary. Users who deleted old Meta LLaMA .pth files can use the migrate-ggml-2023-03-30-pr613.py script to convert old ggml files into the new mmap()'able format.written, reviewed, cited

The new format merges multi-file models like LLaMA 13B into a single file to prevent C++ code from reshaping data during loading.

Fine-tune your own Llama 2 to replace GPT-3.5/4

news.ycombinator.com2023-09-12HN front pageabout the Llama familyHN 955 pts · 181 comments10,023 words

Fine-tuning open-source Llama 2 models can replace GPT-3.5 and GPT-4 for specific tasks, with a fine-tuned Llama 7B matching GPT-4 labels 95% of the time on a test set. While fine-tuning requires 1000+ examples and is less flexible than prompting, a fine-tuned Llama 7B is 50x cheaper than GPT-3.5 per token, costing just $19 to classify 2M recipes versus $23k for GPT-4.written, reviewed, cited

All coverage tracked for this model (14) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
Llama 3.2: Revolutionizing edge AI and vision with open, customizable models	ai.meta.com official	2024-09-25	2026-10-07	hn
Llama.ttf: A font which is also an LLM	fuglede.github.io	2024-06-23	2026-10-07	hn
Show HN: I built a free in-browser Llama 3 chatbot powered by WebGPU	github.com	2024-05-03	2026-10-07	hn
Meta Llama 3	llama.meta.com official	2024-04-18	2026-10-07	hn
Meta AI releases Code Llama 70B	twitter.com	2024-01-29	2026-10-07	hn
Fine-tune your own Llama 2 to replace GPT-3.5/4	news.ycombinator.com	2023-09-12	2026-10-07	hn
Code Llama, a state-of-the-art large language model for coding	ai.meta.com official	2023-08-24	2026-10-07	hn
Run Llama 2 uncensored locally	ollama.ai	2023-08-02	2026-10-07	hn
Llama2.c: Inference llama 2 in one file of pure C	github.com	2023-07-23	2026-10-07	hn
Run Llama 13B with a 6GB graphics card	gist.github.com	2023-05-14	2026-10-07	hn
The LLama Effect: Leak Sparked a Series of Open Source Alternatives to ChatGPT	thesequence.substack.com	2023-04-09	2026-10-07	hn
Llama.cpp 30B runs with only 6GB of RAM now	github.com	2023-03-31	2026-10-07	hn
Gpt4all: A chatbot trained on ~800k GPT-3.5-Turbo Generations based on LLaMa	github.com	2023-03-28	2026-10-07	hn
Llama.cpp: Port of Facebook's LLaMA model in C/C++, with Apple Silicon support	github.com	2023-03-10	2026-10-07	hn

Papers

2 papers

Papers 2

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations this model

Hakan Inan, Kartikeya Upasani, Jianfeng Chi +8 · 2023-12-07 · PDF · HF paper page (9 upvotes)
RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content cited in card

Zhuowen Yuan, Zidi Xiong, Yi Zeng +4 · 2024-03-19 · PDF · HF paper page (3 upvotes)

Similar models

3 models

Similar models

Llama-3.1-8B-Instruct · 8.03B
same size band (M), same task (text-generation), more providers
Qwen3-8B · 8.19B
same size band (M), same task (text-generation), more providers, permissive license
Meta-Llama-3-8B-Instruct · 8.03B
same size band (M), same task (text-generation), more providers

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:41 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:13 UTC · Last verified 2026-10-25

Follow Meta-Llama-Guard-2-8B

Watch this model

/feeds/models/meta-llama--Meta-Llama-Guard-2-8B.xml(RSS 2.0) · every change to Meta-Llama-Guard-2-8B
/feeds/families/meta-llama--Llama.xml(RSS 2.0) · every change across the Llama family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.