panda_guard.cli package

Submodules

panda_guard.cli.attack module

panda_guard.cli.attack.apply_env_vars_to_config(config_dict: Dict[str, Any], model_type: str) → Dict[str, Any]

Apply environment variables to the config if needed.

panda_guard.cli.attack.display_help()

Display available commands.

panda_guard.cli.attack.display_judge_results(results)

Display judge evaluation results in a less intrusive format.

panda_guard.cli.attack.display_token_info(usage, response_time=None)

Display token usage information in a less intrusive format.

panda_guard.cli.attack.get_package_config_path(model_type: str) → Path

Get the path to a default config file within the package.

panda_guard.cli.attack.is_iterator(obj)

Check if an object is an iterator but not a list or other sequence.

panda_guard.cli.attack.load_yaml(yaml_file)

Load YAML configuration file

panda_guard.cli.attack.start(config: str | None = <typer.models.ArgumentInfo object>, attacker: ~pathlib.Path | None = <typer.models.OptionInfo object>, endpoint: ~pathlib.Path | None = <typer.models.OptionInfo object>, model: ~pathlib.Path | None = <typer.models.OptionInfo object>, device: str | None = <typer.models.OptionInfo object>, log_level: str = <typer.models.OptionInfo object>, output: ~pathlib.Path | None = <typer.models.OptionInfo object>, stream: bool = <typer.models.OptionInfo object>, verbose: bool = <typer.models.OptionInfo object>)

Start an interactive attack session using configuration from a YAML file or a predefined model type.

If config is a file path ending with .yaml, it will load configuration from that file. If config is one of ‘openai’, ‘gemini’, or ‘claude’, it will load a default configuration and apply relevant environment variables.

panda_guard.cli.chat module

panda_guard.cli.chat.apply_env_vars_to_config(config_dict: Dict[str, Any], model_type: str) → Dict[str, Any]

Apply environment variables to the config if needed.

panda_guard.cli.chat.display_help()

Display available commands.

panda_guard.cli.chat.display_judge_results(results)

Display judge evaluation results in a less intrusive format.

panda_guard.cli.chat.display_token_info(usage, response_time=None)

Display token usage information in a less intrusive format.

panda_guard.cli.chat.get_package_config_path(model_type: str) → Path

Get the path to a default config file within the package.

panda_guard.cli.chat.is_iterator(obj)

Check if an object is an iterator but not a list or other sequence.

panda_guard.cli.chat.load_yaml(yaml_file)

Load YAML configuration file

panda_guard.cli.chat.start(config: str | None = <typer.models.ArgumentInfo object>, defense: ~pathlib.Path | None = <typer.models.OptionInfo object>, judge: str | None = <typer.models.OptionInfo object>, endpoint: ~pathlib.Path | None = <typer.models.OptionInfo object>, model: ~pathlib.Path | None = <typer.models.OptionInfo object>, temperature: float | None = <typer.models.OptionInfo object>, device: str | None = <typer.models.OptionInfo object>, log_level: str = <typer.models.OptionInfo object>, output: ~pathlib.Path | None = <typer.models.OptionInfo object>, stream: bool = <typer.models.OptionInfo object>, verbose: bool = <typer.models.OptionInfo object>)

Start an interactive chat session using configuration from a YAML file or a predefined model type.

If config is a file path ending with .yaml, it will load configuration from that file. If config is one of ‘openai’, ‘gemini’, or ‘claude’, it will load a default configuration and apply relevant environment variables.

panda_guard.cli.eval module

panda_guard.cli.eval.fill_llms_configs(d, parent_key='', llm_configs=None)

panda_guard.cli.eval.get_input_files(input_dir)

panda_guard.cli.eval.get_package_config_path(model_type: str) → Path

Get the path to a default config file within the package.

panda_guard.cli.eval.is_iterator(obj)

Check if an object is an iterator but not a list or other sequence.

panda_guard.cli.eval.load_json_files_from_yaml(yaml_file)

panda_guard.cli.eval.load_yaml(yaml_file)

Load YAML configuration file

panda_guard.cli.eval.override_config(config_dict)

panda_guard.cli.eval.process_file(json_file, input_dir, output_dir, attacker_config, defender_config, judge_configs, config_dict)

panda_guard.cli.eval.run_inference(pipe, messages, goal)

panda_guard.cli.eval.start(config: str | None = <typer.models.ArgumentInfo object>, input_dir: str = <typer.models.OptionInfo object>, output_dir: str = <typer.models.OptionInfo object>, log_level: str = <typer.models.OptionInfo object>, num_workers: int = <typer.models.OptionInfo object>)

Start an eval using configuration from a YAML file or a predefined model type.

If config is a file path ending with .yaml, it will load configuration from that file. If config is one of ‘openai’, ‘gemini’, or ‘claude’, it will load a default configuration and apply relevant environment variables.

panda_guard.cli.inference module

panda_guard.cli.inference.apply_env_vars_to_config(config_dict: Dict[str, Any], model_type: str) → Dict[str, Any]

Apply environment variables to the config if needed.

panda_guard.cli.inference.display_judge_results(results)

Display judge evaluation results in a less intrusive format.

panda_guard.cli.inference.display_token_info(usage, response_time=None)

Display token usage information in a less intrusive format.

panda_guard.cli.inference.get_package_config_path(model_type: str) → Path

Get the path to a default config file within the package.

panda_guard.cli.inference.is_iterator(obj)

Check if an object is an iterator but not a list or other sequence.

panda_guard.cli.inference.load_yaml(yaml_file)

Load YAML configuration file

panda_guard.cli.inference.start(config: str | None = <typer.models.ArgumentInfo object>, input_file: str | None = <typer.models.OptionInfo object>, attack: ~pathlib.Path | None = <typer.models.OptionInfo object>, defense: ~pathlib.Path | None = <typer.models.OptionInfo object>, endpoint: ~pathlib.Path | None = <typer.models.OptionInfo object>, model: ~pathlib.Path | None = <typer.models.OptionInfo object>, device: str | None = <typer.models.OptionInfo object>, log_level: str = <typer.models.OptionInfo object>, output_dir: str = <typer.models.OptionInfo object>, verbose: bool = <typer.models.OptionInfo object>)

Start an inference session using configuration from a YAML file or a predefined model type.

If config is a file path ending with .yaml, it will load configuration from that file. If config is one of ‘openai’, ‘gemini’, or ‘claude’, it will load a default configuration and apply relevant environment variables.

panda_guard.cli.main module

panda_guard.cli.main.main(version: bool = <typer.models.OptionInfo object>)

Panda Guard CLI for chatting with and serving language models.

panda_guard.cli.main.version_callback(value: bool)

panda_guard.cli.serve module

class panda_guard.cli.serve.ChatCompletionRequest(*, model: str, messages: List[Dict[str, str]], temperature: float | None = 1.0, top_p: float | None = 1.0, n: int | None = 1, stream: bool | None = False, max_tokens: int | None = None, presence_penalty: float | None = 0.0, frequency_penalty: float | None = 0.0, user: str | None = None)

Bases: BaseModel

frequency_penalty: float | None

max_tokens: int | None

messages: List[Dict[str, str]]

model: str

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

n: int | None

presence_penalty: float | None

stream: bool | None

temperature: float | None

top_p: float | None

user: str | None

class panda_guard.cli.serve.ChatCompletionResponse(*, id: str, object: str = 'chat.completion', created: int, model: str, choices: List[Choice], usage: Dict[str, int])

Bases: BaseModel

choices: List[Choice]

created: int

id: str

model: str

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

object: str

usage: Dict[str, int]

class panda_guard.cli.serve.Choice(*, index: int, message: Dict[str, str], finish_reason: str)

Bases: BaseModel

finish_reason: str

index: int

message: Dict[str, str]

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

class panda_guard.cli.serve.ModelData(*, id: str, object: str = 'model', created: int, owned_by: str = 'panda-guard')

Bases: BaseModel

created: int

id: str

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

object: str

owned_by: str

class panda_guard.cli.serve.ModelsResponse(*, object: str = 'list', data: List[ModelData])

Bases: BaseModel

data: List[ModelData]

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

object: str

panda_guard.cli.serve.apply_env_vars_to_config(config_dict: Dict[str, Any], model_type: str) → Dict[str, Any]

Apply environment variables to the config if needed.

panda_guard.cli.serve.create_fastapi_app(pipeline_instance)

Create a FastAPI application with OpenAI-compatible endpoints.

panda_guard.cli.serve.get_package_config_path(model_type: str) → Path

Get the path to a default config file within the package.

panda_guard.cli.serve.load_yaml(yaml_file)

Load YAML configuration file

panda_guard.cli.serve.start(config: str | None = <typer.models.ArgumentInfo object>, defense: ~pathlib.Path | None = <typer.models.OptionInfo object>, judge: str | None = <typer.models.OptionInfo object>, endpoint: ~pathlib.Path | None = <typer.models.OptionInfo object>, model: ~pathlib.Path | None = <typer.models.OptionInfo object>, temperature: float | None = <typer.models.OptionInfo object>, device: str | None = <typer.models.OptionInfo object>, log_level: str = <typer.models.OptionInfo object>, port: int = <typer.models.OptionInfo object>, host: str = <typer.models.OptionInfo object>, verbose: bool = <typer.models.OptionInfo object>)

Start an API server compatible with OpenAI API format.

Accepts the same configuration options as the chat interface, plus host and port settings.

Module contents