Hugging Face's logo

LiteLLMs
/
Meta-Llama-Guard-2-8B-GGUF

Text Generation
GGUF
PyTorch
English
facebook
meta
llama
llama-3
GGUF
Model card Files Community
Edit model card

Meta-Llama-Guard-2-8B-GGUF

Description

This repo contains GGUF format model files for Meta-Llama-Guard-2-8B.

About GGUF

GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp. Here is an incomplete list of clients and libraries that are known to support GGUF:

  • llama.cpp. This is the source project for GGUF, providing both a Command Line Interface (CLI) and a server option.
  • text-generation-webui, Known as the most widely used web UI, this project boasts numerous features and powerful extensions, and supports GPU acceleration.
  • Ollama Ollama is a lightweight and extensible framework designed for building and running language models locally. It features a simple API for creating, managing, and executing models, along with a library of pre-built models for use in various applications​
  • KoboldCpp, A comprehensive web UI offering GPU acceleration across all platforms and architectures, particularly renowned for storytelling.
  • GPT4All, This is a free and open source GUI that runs locally, supporting Windows, Linux, and macOS with full GPU acceleration.
  • LM Studio An intuitive and powerful local GUI for Windows and macOS (Silicon), featuring GPU acceleration.
  • LoLLMS Web UI. A notable web UI with a variety of unique features, including a comprehensive model library for easy model selection.
  • Faraday.dev, An attractive, user-friendly character-based chat GUI for Windows and macOS (both Silicon and Intel), also offering GPU acceleration.
  • llama-cpp-python, A Python library equipped with GPU acceleration, LangChain support, and an OpenAI-compatible API server.
  • candle, A Rust-based ML framework focusing on performance, including GPU support, and designed for ease of use.
  • ctransformers, A Python library featuring GPU acceleration, LangChain support, and an OpenAI-compatible AI server.
  • localGPT An open-source initiative enabling private conversations with documents.

Explanation of quantisation methods

Click to see details The new methods available are:
  • GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
  • GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
  • GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
  • GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
  • GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw.

How to download GGUF files

Note for manual downloaders: You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single folder.

The following clients/libraries will automatically download models for you, providing a list of available models to choose from:

  • LM Studio
  • LoLLMS Web UI
  • Faraday.dev

In text-generation-webui

Under Download Model, you can enter the model repo: LiteLLMs/Meta-Llama-Guard-2-8B-GGUF and below it, a specific filename to download, such as: Q4_0/Q4_0-00001-of-00009.gguf.

Then click Download.

On the command line, including multiple files at once

I recommend using the huggingface-hub Python library:

pip3 install huggingface-hub

Then you can download any individual model file to the current directory, at high speed, with a command like this:

huggingface-cli download LiteLLMs/Meta-Llama-Guard-2-8B-GGUF Q4_0/Q4_0-00001-of-00009.gguf --local-dir . --local-dir-use-symlinks False
More advanced huggingface-cli download usage (click to read)

You can also download multiple files at once with a pattern:

huggingface-cli download LiteLLMs/Meta-Llama-Guard-2-8B-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'

For more documentation on downloading with huggingface-cli, please see: HF -> Hub Python Library -> Download files -> Download from the CLI.

To accelerate downloads on fast connections (1Gbit/s or higher), install hf_transfer:

pip3 install huggingface_hub[hf_transfer]

And set environment variable HF_HUB_ENABLE_HF_TRANSFER to 1:

HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download LiteLLMs/Meta-Llama-Guard-2-8B-GGUF Q4_0/Q4_0-00001-of-00009.gguf --local-dir . --local-dir-use-symlinks False

Windows Command Line users: You can set the environment variable by running set HF_HUB_ENABLE_HF_TRANSFER=1 before the download command.

## Example `llama.cpp` command

Make sure you are using llama.cpp from commit d0cee0d or later.

./main -ngl 35 -m Q4_0/Q4_0-00001-of-00009.gguf --color -c 8192 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<PROMPT>"

Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.

Change -c 8192 to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically. Note that longer sequence lengths require much more resources, so you may need to reduce this value.

If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins

For other parameters and how to use them, please refer to the llama.cpp documentation

How to run in text-generation-webui

Further instructions can be found in the text-generation-webui documentation, here: text-generation-webui/docs/04 ‐ Model Tab.md.

How to run from Python code

You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.

How to load this model in Python code, using llama-cpp-python

For full documentation, please see: llama-cpp-python docs.

First install the package

Run one of the following commands, according to your system:

# Base ctransformers with no GPU acceleration
pip install llama-cpp-python
# With NVidia CUDA acceleration
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
# Or with OpenBLAS acceleration
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
# Or with CLBLast acceleration
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
# Or with AMD ROCm GPU acceleration (Linux only)
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
# Or with Metal GPU acceleration for macOS systems only
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
pip install llama-cpp-python

Simple llama-cpp-python example code 

from llama_cpp import Llama
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = Llama(
  model_path="./Q4_0/Q4_0-00001-of-00009.gguf",  # Download the model file first
  n_ctx=32768,  # The max sequence length to use - note that longer sequence lengths require much more resources
  n_threads=8,            # The number of CPU threads to use, tailor to your system and the resulting performance
  n_gpu_layers=35         # The number of layers to offload to GPU, if you have GPU acceleration available
)
# Simple inference example
output = llm(
  "<PROMPT>", # Prompt
  max_tokens=512,  # Generate up to 512 tokens
  stop=["</s>"],   # Example stop token - not necessarily correct for this specific model! Please check before using.
  echo=True        # Whether to echo the prompt
)
# Chat Completion API
llm = Llama(model_path="./Q4_0/Q4_0-00001-of-00009.gguf", chat_format="llama-2")  # Set chat_format according to the model you are using
llm.create_chat_completion(
    messages = [
        {"role": "system", "content": "You are a story writing assistant."},
        {
            "role": "user",
            "content": "Write a story about llamas."
        }
    ]
)

How to use with LangChain

Here are guides on using llama-cpp-python and ctransformers with LangChain:

Original model card: Meta-Llama-Guard-2-8B

Model Details

Meta Llama Guard 2 is an 8B parameter Llama 3-based [1] LLM safeguard model. Similar to Llama Guard, it can be used for classifying content in both LLM inputs (prompt classification) and in LLM responses (response classification). It acts as an LLM – it generates text in its output that indicates whether a given prompt or response is safe or unsafe, and if unsafe, it also lists the content categories violated. Below is a response classification example input and output for Llama Guard 2.

In order to produce classifier scores, we look at the probability for the first token, and use that as the “unsafe” class probability. We can then apply score thresholding to make binary decisions.

Harm Taxonomy and Policy

The model is trained to predict safety labels on the 11 categories shown below, based on the MLCommons taxonomy of hazards.

Harm categories
S1: Violent Crimes S2: Non-Violent Crimes
S3: Sex-Related Crimes S4: Child Sexual Exploitation
S5: Specialized Advice S6: Privacy
S7: Intellectual Property S8: Indiscriminate Weapons
S9: Hate S10: Suicide & Self-Harm
S11: Sexual Content

S1: Violent Crimes

Responses that enable, encourage, or endorse the commission of violent crimes, including: (1) unlawful violence toward people (ex: terrorism, genocide, murder, hate-motivated violent crimes, child abuse, assault, battery, kidnapping) and (2) unlawful violence toward animals (ex: animal abuse)

S2: Non-Violent Crimes

Responses that enable, encourage, or endorse the commission of non-violent crimes, including: (1) personal crimes (ex: labor trafficking, threats, intimidation, hate-motivated non-violent crimes), (2) financial crimes (ex: fraud, scams, money laundering), (3) property crimes (ex: theft, arson, vandalism), (4) drug crimes (ex: creating or using regulated substances), (5) weapons crimes (ex: producing unlicensed firearms), and (6) cyber crimes (ex: hacking)

S3: Sex-Related Crimes

Responses that enable, encourage, or endorse the commission of sex-related crimes, including: (1) sex trafficking, (2) sexual assault (ex: rape), (3) sexual harassment (physical, verbal, or visual), and (4) prostitution

S4: Child Sexual Exploitation

Responses that contain, describe, enable, encourage, or endorse the sexual abuse of children

S5: Specialized Advice

Responses that contain specialized financial, medical, or legal advice, or that indicate dangerous activities or objects are safe

S6: Privacy

Responses that contain sensitive, nonpublic personal information that could undermine someone's physical, digital, or financial security

S7: Intellectual Property

Responses that may violate the intellectual property rights of any third party

S8: Indiscriminate Weapons

Responses that enable, encourage, or endorse the creation of indiscriminate weapons, including: (1) chemical weapons (ex: nerve gas), (2) biological weapons (ex: anthrax), (3) radiological weapons (ex: cobalt bombs), (4) nuclear weapons (ex: fission bombs), and (5) high-yield explosive weapons (ex: cluster munitions)

S9: Hate

Responses that demean or dehumanize people on the basis of their sensitive, personal characteristics (i.e., race, color, ethnicity, national origin, disability, religious affiliation, caste, sexual orientation, sex, gender identity, and/or serious disease)

S10: Suicide & Self-Harm

Responses that enable, encourage, or endorse acts of intentional self-harm, including: (1) suicide, (2) self-injury (ex: cutting), and (3) disordered eating

S11: Sexual Content

Responses that contain erotica

How to Use in transformers 

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "meta-llama/Meta-Llama-Guard-2-8B"
device = "cuda"
dtype = torch.bfloat16

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=dtype, device_map=device)

def moderate(chat):
    input_ids = tokenizer.apply_chat_template(chat, return_tensors="pt").to(device)
    output = model.generate(input_ids=input_ids, max_new_tokens=100, pad_token_id=0)
    prompt_len = input_ids.shape[-1]
    return tokenizer.decode(output[0][prompt_len:], skip_special_tokens=True)

moderate([
    {"role": "user", "content": "I forgot how to kill a process in Linux, can you help?"},
    {"role": "assistant", "content": "Sure! To kill a process in Linux, you can use the kill command followed by the process ID (PID) of the process you want to terminate."},
])
# `safe`

Training and Evaluation Data

We start with the training set of Llama Guard (see model card), and obtain labels on the Harm Taxonomy described above. To improve adaptability of the model to different prompts, we train on hard samples, which are obtained by taking an existing sample and prompting Llama2 70B to produce an alternate policy description that will flip the label of the given sample.

We report metrics for various models and APIs on our validation set, which is obtained from combining the validation set of Llama Guard v1 and held-out samples from the additional Llama 3 safety data.

We compare performance on our internal test set, as well as on open datasets like XSTest, OpenAI moderation, and BeaverTails.

We find that there is overlap between our training set and the BeaverTails-30k test split. Since both our internal test set and BeaverTails use prompts from the Anthropic's hh-rlhf dataset as a starting point for curating data, it is possible that different splits of Anthropic were used while creating the two datasets. Therefore to prevent leakage of signal between our train set and the BeaverTails-30k test set, we create our own BeaverTails-30k splits based on the Anthropic train-test splits used for creating our internal sets.

Note on evaluations: As discussed in the Llama Guard paper, comparing model performance is not straightforward as each model is built on its own policy and is expected to perform better on an evaluation dataset with a policy aligned to the model. This highlights the need for industry standards. By aligning Llama Guard 2 with the Proof of Concept MLCommons taxonomy, we hope to drive adoption of industry standards like this and facilitate collaboration and transparency in the LLM safety and content evaluation space.

Model Performance

We evaluate the performance of Llama Guard 2 and compare it with Llama Guard and popular content moderation APIs such as Azure, OpenAI Moderation, and Perspective. We use the token probability of the first output token (i.e. safe/unsafe) as the score for classification. For obtaining a binary classification decision from the score, we use a threshold of 0.5.

Llama Guard 2 improves over Llama Guard, and outperforms other approaches on our internal test set. Note that we manage to achieve great performance while keeping a low false positive rate as we know that over-moderation can impact user experience when building LLM-applications.

| Model | F1 ↑ | AUPRC ↑ | False Positive
Rate ↓ | | | :--: | :--: | :--: | | Llama Guard* | 0.665 | 0.854 | 0.027 | | Llama Guard 2 | 0.915 | 0.974 | 0.040 | | GPT4 | 0.796 | N/A | 0.151 | | OpenAI Moderation API | 0.347 | 0.669 | 0.030 | | Azure Content Safety API | 0.519 | N/A | 0.245 | | Perspective API | 0.265 | 0.586 | 0.046 |

Table 1: Comparison of performance of various approaches measured on our internal test set.
*The performance of Llama Guard is lower on our new test set due to expansion of the number of harm categories from 6 to 11, which is not aligned to what Llama Guard was trained on.


| Category | False Negative Rate* ↓ | False Positive Rate ↓ | | --: | | Violent Crimes | 0.042 | 0.002 | | Privacy | 0.057 | 0.004 | | Non-Violent Crimes | 0.082 | 0.009 | | Intellectual Property | 0.099 | 0.004 | | Hate | 0.190 | 0.005 | | Specialized Advice | 0.192 | 0.009 | | Sexual Content | 0.229 | 0.004 | | Indiscriminate Weapons | 0.263 | 0.001 | | Child Exploitation | 0.267 | 0.000 | | Sex Crimes | 0.275 | 0.002 | | Self-Harm | 0.277 | 0.002 |

Table 2: Category-wise breakdown of false negative rate and false positive rate for Llama Guard 2 on our internal benchmark for response classification with safety labels from the ML Commons taxonomy.
*The binary safe/unsafe label is used to compute categorical FNR by using the true categories. We do not penalize the model while computing FNR for cases where the model predicts the correct overall label but an incorrect categorical label.

We also report performance on OSS safety datasets, though we note that the policy used for assigning safety labels is not aligned with the policy used while training Llama Guard 2. Still, Llama Guard 2 provides a superior tradeoff between f1 score and False Positive Rate on the XSTest and OpenAI Moderation datasets, demonstrating good adaptability to other policies.

The BeaverTails dataset has a lower bar for a sample to be considered unsafe compared to Llama Guard 2's policy. The policy and training data of MDJudge [4] is more aligned with this dataset and we see that it performs better on them as expected (at the cost of a higher FPR). GPT-4 achieves high recall on all of the sets but at the cost of very high FPR (9-25%), which could hurt its ability to be used as a safeguard for practical applications.

(F1 ↑ / False Positive Rate ↓)
False Refusals
(XSTest)
OpenAI policy
(OpenAI Mod)
BeaverTails policy
(BeaverTails-30k)
Llama Guard 0.737 / 0.079 0.737 / 0.079 0.599 / 0.035
Llama Guard 2 0.884 / 0.084 0.807 / 0.060 0.736 / 0.059
MDJudge 0.856 / 0.172 0.768 / 0.212 0.849 / 0.098
GPT4 0.895 / 0.128 0.842 / 0.092 0.802 / 0.256
OpenAI Mod API 0.576 / 0.040 0.788 / 0.156 0.284 / 0.056
Table 3: Comparison of performance of various approaches measured on our internal test set for response classification.
NOTE: The policy used for training Llama Guard does not align with those used for labeling these datasets. Still, Llama Guard 2 provides a superior tradeoff between F1 score and False Positive Rate across these datasets, demonstrating strong adaptability to other policies.

We hope to provide developers with a high-performing moderation solution for most use cases by aligning Llama Guard 2 taxonomy with MLCommons standard. But as outlined in our Responsible Use Guide, each use case requires specific safety considerations and we encourage developers to tune Llama Guard 2 for their own use case to achieve better moderation for their custom policies. As an example of how Llama Guard 2's performance may change, we train on the BeaverTails training dataset and compare against MDJudge (which was trained on BeaverTails among others).

Model F1 ↑ False Positive Rate ↓
Llama Guard 2 0.736 0.059
MDJudge 0.849 0.098
Llama Guard 2 + BeaverTails 0.852 0.101

Table 4: Comparison of performance on BeaverTails-30k.

Limitations

There are some limitations associated with Llama Guard 2. First, Llama Guard 2 itself is an LLM fine-tuned on Llama 3. Thus, its performance (e.g., judgments that need common sense knowledge, multilingual capability, and policy coverage) might be limited by its (pre-)training data.

Second, Llama Guard 2 is finetuned for safety classification only (i.e. to generate "safe" or "unsafe"), and is not designed for chat use cases. However, since it is an LLM, it can still be prompted with any text to obtain a completion.

Lastly, as an LLM, Llama Guard 2 may be susceptible to adversarial attacks or prompt injection attacks that could bypass or alter its intended use. However, with the help of external components (e.g., KNN, perplexity filter), recent work (e.g., [3]) demonstrates that Llama Guard is able to detect harmful content reliably.

Note on Llama Guard 2's policy

Llama Guard 2 supports 11 out of the 13 categories included in the MLCommons AI Safety taxonomy. The Election and Defamation categories are not addressed by Llama Guard 2 as moderating these harm categories requires access to up-to-date, factual information sources and the ability to determine the veracity of a particular output. To support the additional categories, we recommend using other solutions (e.g. Retrieval Augmented Generation) in tandem with Llama Guard 2 to evaluate information correctness.

Citation 

@misc{metallamaguard2,
  author =       {Llama Team},
  title =        {Meta Llama Guard 2},
  howpublished = {\url{https://github.com/meta-llama/PurpleLlama/blob/main/Llama-Guard2/MODEL_CARD.md}},
  year =         {2024}
}

References

[1] Llama 3 Model Card

[2] Llama Guard Model Card

[3] RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content

[4] MDJudge for Salad-Bench

Downloads last month
399
GGUF
Model size
8.03B params
Architecture
llama
View files
Inference Examples
Text Generation
Unable to determine this model's library. Check the docs .