Source code for pyrit.executor.promptgen.fuzzer.fuzzer_converter_base
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.
import json
import logging
import uuid
from pyrit.common.apply_defaults import REQUIRED_VALUE, apply_defaults
from pyrit.exceptions import (
InvalidJsonException,
pyrit_json_retry,
remove_markdown_json,
)
from pyrit.models import (
Message,
MessagePiece,
PromptDataType,
SeedPrompt,
)
from pyrit.prompt_converter import ConverterResult, PromptConverter
from pyrit.prompt_target import PromptChatTarget
logger = logging.getLogger(__name__)
[docs]
class FuzzerConverter(PromptConverter):
"""
Base class for GPTFUZZER converters.
Adapted from GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts.
Paper: https://arxiv.org/pdf/2309.10253 by Jiahao Yu, Xingwei Lin, Zheng Yu, Xinyu Xing.
GitHub: https://github.com/sherdencooper/GPTFuzz/tree/master
"""
[docs]
@apply_defaults
def __init__(
self,
*,
converter_target: PromptChatTarget = REQUIRED_VALUE, # type: ignore[assignment]
prompt_template: SeedPrompt,
):
"""
Initialize the converter with the specified chat target and prompt template.
Args:
converter_target (PromptChatTarget): Chat target used to perform fuzzing on user prompt.
Can be omitted if a default has been configured via PyRIT initialization.
prompt_template (SeedPrompt): Template to be used instead of the default system prompt with
instructions for the chat target.
Raises:
ValueError: If converter_target is not provided and no default has been configured.
"""
self.converter_target = converter_target
self.system_prompt = prompt_template.value
self.template_label = "TEMPLATE"
[docs]
def update(self, **kwargs) -> None:
"""Update the converter with new parameters."""
pass
[docs]
async def convert_async(self, *, prompt: str, input_type: PromptDataType = "text") -> ConverterResult:
"""
Convert the given prompt into the target format supported by the converter.
Args:
prompt (str): The prompt to be converted.
input_type (PromptDataType): The type of input data.
Returns:
ConverterResult: The result containing the modified prompt.
Raises:
ValueError: If the input type is not supported.
"""
if not self.input_supported(input_type):
raise ValueError("Input type not supported")
conversation_id = str(uuid.uuid4())
self.converter_target.set_system_prompt(
system_prompt=self.system_prompt,
conversation_id=conversation_id,
attack_identifier=None,
)
formatted_prompt = f"===={self.template_label} BEGINS====\n{prompt}\n===={self.template_label} ENDS===="
prompt_metadata: dict[str, str | int] = {"response_format": "json"}
request = Message(
[
MessagePiece(
role="user",
original_value=formatted_prompt,
converted_value=formatted_prompt,
conversation_id=conversation_id,
sequence=1,
prompt_target_identifier=self.converter_target.get_identifier(),
original_value_data_type=input_type,
converted_value_data_type=input_type,
converter_identifiers=[self.get_identifier()],
prompt_metadata=prompt_metadata,
)
]
)
response = await self.send_prompt_async(request)
return ConverterResult(output_text=response, output_type="text")
[docs]
@pyrit_json_retry
async def send_prompt_async(self, request):
"""
Send the message to the converter target and process the response.
Args:
request: The message request to send.
Returns:
str: The output from the parsed JSON response.
Raises:
InvalidJsonException: If the response is not valid JSON or missing required keys.
"""
response = await self.converter_target.send_prompt_async(message=request)
response_msg = response[0].get_value()
response_msg = remove_markdown_json(response_msg)
try:
parsed_response = json.loads(response_msg)
if "output" not in parsed_response:
raise InvalidJsonException(message=f"Invalid JSON encountered; missing 'output' key: {response_msg}")
return parsed_response["output"]
except json.JSONDecodeError:
raise InvalidJsonException(message=f"Invalid JSON encountered: {response_msg}")
[docs]
def output_supported(self, output_type: PromptDataType) -> bool:
"""
Check if the output type is supported.
Returns:
bool: True if output type is text, False otherwise.
"""
return output_type == "text"
