garak
is a modular tool for probing LLMs for undesireable prompt responses
invoke with python3 -m garak
currently supports:
- hugging face hub generative models
- replicate text models
- openai api chat & continuation models
- ggml models like llama.cpp
$ python3 -m garak --help
garak LLM probe v0.9 ( https://github.com/leondz/garak ) at 2023-06-08T22:49:32.650003
usage: __main__.py [-h] [--model_type MODEL_TYPE] [--model_name MODEL_NAME] [--seed [SEED]] [--generations GENERATIONS] [--probes PROBES] [--detectors DETECTORS]
[--eval_threshold EVAL_THRESHOLD] [--plugin_info PLUGIN_INFO] [--list_probes] [--list_detectors] [--list_generators] [-V] [-v]
LLM safety & security scanning tool
options:
-h, --help show this help message and exit
--model_type MODEL_TYPE, -m MODEL_TYPE
module and optionally also class of the generator, e.g. 'huggingface', or 'openai'
--model_name MODEL_NAME, -n MODEL_NAME
name of the model, e.g. 'timdettmers/guanaco-33b-merged'
--seed [SEED], -s [SEED]
random seed
--generations GENERATIONS, -g GENERATIONS
number of generations per prompt
--probes PROBES, -p PROBES
list of probe names to use, or 'all' for all (default).
--detectors DETECTORS, -d DETECTORS
list of detectors to use, or 'all' for all. Default is to use the probe's suggestion.
--eval_threshold EVAL_THRESHOLD
minimum threshold for a successful hit
--plugin_info PLUGIN_INFO, -P PLUGIN_INFO
show info about one plugin
--list_probes list available vulnerability probes
--list_detectors list available detectors
--list_generators list available generation model interfaces
-V, --version print version info & exit
-v, --verbose
See https://github.com/leondz/garak
garak
is a command-line tool. It's developed in Linux and OSX.
Just grab it from PyPI and you should be good to go:
python -m pip install -U garak
The standard pip version of garak
is updated periodically. To get a fresher version, from GitHub, try:
python -m pip install -U git+https://github.com/leondz/garak.git@main
garak
has its own dependencies. You can to install garak
in its own Conda environment:
conda create --name garak python>=3.9
conda activate garak
gh repo clone leondz/garak
cd garak
python -m pip install -r requirements.txt
OK, if that went fine, you're probably good to go!
The general syntax is:
python3 -m garak <options>
garak
needs to know what model to scan, and by default, it'll try all the probes it knows on that model, using the vulnerability detectors recommended by each probe. You can see a list of probes using:
python3 -m garak --list_probes
To specify a generator, use the --model_name
and, optionally, the --model_type
options. Model name specifies a model family/interface; model type specifies the exact model to be used. The "Intro to generators" section below describes some of the generators supported. A straightfoward generator family is Hugging Face models; to load one of these, set --model_name
to huggingface
and --model_type
to the model's name on Hub (e.g. "RWKV/rwkv-4-169m-pile"
). Some generators might need an API key to be set as an environment variable, and they'll let you know if they need that.
garak
runs all the probes by default, but you can be specific about that too. --probes promptinject
will use only the PromptInject framework's methods, for example. You can also specify one specific plugin instead of a plugin family by adding the plugin name after a .
; for example, --probes lmrc.SlurUsage
will use an implementation of checking for models generating slurs based on the Language Model Risk Cards framework.
For help & inspiration, find us on twitter or discord!
Probe ChatGPT for encoding-based prompt injection (OSX/*nix) (replace example value with a real OpenAI API key)
export OPENAI_API_KEY="sk-123XXXXXXXXXXXX"
python3 -m garak --model_type openai --model_name gpt-3.5-turbo --probes encoding
See if the Hugging Face version of GPT2 is vulnerable to DAN 11.0
python3 -m garak --model_type huggingface --model_name gpt2 --probes dan.Dan_11_0
For each probe loaded, garak will print a progress bar as it generates. Once generation is complete, a row evaluating that probe's results on each detector is given. If any of the prompt attempts yielded an undesirable behaviour, the response will be marked as FAIL, and the failure rate given.
Here are the results with the encoding
module on a GPT-3 variant:
And the same results for ChatGPT:
We can see that the more recent model is much more susceptible to encoding-based injection attacks, where text-babbage-001 was only found to be vulnerable to quoted-printable and MIME encoding injections. The figures at the end of each row, e.g. 840/840, indicate the number of text generations total and then how many of these seemed to behave OK. The figure can be quite high because more than one generation is made per prompt - by default, 10.
Errors go in garak.log
; the run is logged in detail in a .jsonl
file specified at analysis start & end. There's a basic analysis script in analyse/analyse_log.py
which will output the probes and prompts that led to the most hits.
Send PRs & open issues. Happy hunting!
-
--model_name huggingface
(for transformers models to run locally) -
--model_type
- use the model name from Hub. Only generative models will work. If it fails and shouldn't, please open an issue and paste in the command you tried + the exception! -
--model_name huggingface.InferenceAPI
(for API-based model access) -
--model_type
- the model name from Hub, e.g."mosaicml/mpt-7b-instruct"
-
(optional) set the
HF_INFERENCE_TOKEN
environment variable to a Hugging Face API token with the "read" role; see https://huggingface.co/settings/tokens when logged in
--model_name openai
--model_type
- the OpenAI model you'd like to use.text-babbage-001
is fast and fine for testing;gpt-4
seems weaker to many of the more subtle attacks.- set the
OPENAI_API_KEY
environment variable to your OpenAI API key (e.g. "sk-19763ASDF87q6657"); see https://platform.openai.com/account/api-keys when logged in
Recognised model types are whitelisted, because the plugin needs to know which sub-API to use. Completion or ChatCompletion models are OK. If you'd like to use a model not supported, you should get an informative error message, and please send a PR / open an issue.
--model_name replicate
--model_type
- the Replicate model name and hash, e.g."stability-ai/stablelm-tuned-alpha-7b:c49dae36"
- set the
REPLICATE_API_TOKEN
environment variable to your Replicate API token, e.g. "r8-123XXXXXXXXXXXX"; see https://replicate.com/account/api-tokens when logged in
--model_name cohere
--model_type
(optional,command
by default) - The specific Cohere model you'd like to test- set the
COHERE_API_KEY
environment variable to your Cohere API key, e.g. "aBcDeFgHiJ123456789"; see https://dashboard.cohere.ai/api-keys when logged in
--model_name ggml
--model_type
- The path to the ggml model you'd like to load, e.g./home/leon/llama.cpp/models/7B/ggml-model-q4_0.bin
- set the
GGML_MAIN_PATH
environment variable to the path to your ggmlmain
executable
-
--model_name test
-
(alternatively)
--model_name test.Blank
For testing. This always generates the empty string, using thetest.Blank
generator. Will be marked as failing for any tests that require an output, e.g. those that make contentious claims and expect the model to refute them in order to pass. -
--model_name test.Repeat
For testing. This generator repeats back the prompt it received.
A simple probe that always sends an empty prompt.
Probes that test if the model will continue a probably undesirable word
Various DAN and DAN-like attacks
Prompt injection through text encoding
Attempts to have the model generate code for building malware
Probes that attempt to make the model output malicious content signatures
Subsample of the Language Model Risk Cards probes
Attempts to make a model support misleading and false claims
Implementation of the Agency Enterprise PromptInject work (best paper awards @ NeurIPS ML Safety Workshop 2022)
Subset of the RealToxicityPrompts work (data constrained because the full test will take so long to run)
Snowballed Hallucination probes designed to make a model give a wrong answer to questions too complex for it to process
Auto Red-Team. A red-teaming LLM probes the target and reacts to it in an attempt to get toxic output. Prototype, mostly stateless, for now uses a simple GPT-2 fine-tuned on the subset of hhrlhf attempts that yielded detectable toxicity (the only target currently supported for now).
garak
generates two kinds of log:
- A log file,
garak.log
. This includes debugging information fromgarak
and its plugins, and is continued across runs. - A report of the current run, structured as JSONL. A new report file is created every time
garak
runs. The name of this file is output at the beginning and, if successful, also the end of the run. In the report, an entry is made for each probing attempt both as the generations are received, and again when they are evaluated; the entry'sstatus
attribute takes a constant fromgarak.attempts
to describe what stage it was made at.
In a typical run, garak
will read a model type (and optionally model name) from the command line, then determine which probe
s and detector
s to run, start up a generator
, and then pass these to a harness
to do the probing; an evaluator
deals with the results. There are many modules in each of these categories, and each module provides a number of classes that act as individual plugins.
garak/probes/
- classes for generating interactions with LLMsgarak/detectors/
- classes for detecting an LLM is exhibiting a given failure modegarak/evaluators/
- assessment reporting schemesgarak/generators/
- plugins for LLMs to be probedgarak/harnesses/
- classes for structuring testingresources/
- ancillary items required by plugins
The default operating mode is to use the probewise
harness. Given a list of probe module names and probe plugin names, the probewise
harness instantiates each probe, then for each probe reads its recommended_detectors
attribute to get a list of detector
s to run on the output.
Each plugin category (probes
, detectors
, evaluators
, generators
, harnesses
) includes a base.py
which defines the base classes usable by plugins in that category. Each plugin module defines plugin classes that inherit from one of the base classes. For example, garak.generators.openai.OpenAIGenerator
descends from garak.generators.base.Generator
.
Larger artefacts, like model files and bigger corpora, are kept out of the repository; they can be stored on e.g. Hugging Face Hub and loaded locally by clients using garak
.
- Take a look at how other plugins do it
- Inherit from one of the base classes, e.g.
garak.probes.base.TextProbe
- Override as little as possible
- You can test the new code in at least two ways:
- Start an interactive Python session
- Import the model, e.g.
import garak.probes.mymodule
- Instantiate the plugin, e.g.
p = garak.probes.mymodule.MyProbe()
- Import the model, e.g.
- Run a scan with test plugins
- For probes, try a blank generator and always.Pass detector:
python3 -m garak -m test.Blank -p mymodule -d always.Pass
- For detectors, try a blank generator and a blank probe:
python3 -m garak -m test.Blank -p blank -d mymodule
- For generators, try a blank probe and always.Pass detector:
python3 -m garak -m mymodule -p blank -d always.Pass
- For probes, try a blank generator and always.Pass detector:
- Get
garak
to list all the plugins of the type you're writing, with--list_probes
,--list_detectors
, or--list_generators
- Start an interactive Python session
We have an FAQ here. Reach out if you have any more questions!
"Never tell the same lie twice" - Elim
For updates and news see @garak_llm
© GPLv3 2023 Leon Derczynski