hallucination with log probs (#281)

* first init

* fix

* fix test

* new implemenetation

* fix bug

* fix bug

* fix bug

* address issue

* address issues

* address comments

* fix test

* fix

* move constatns

* remove consts
This commit is contained in:
CTran 2024-11-27 15:17:02 -08:00 committed by GitHub
parent f5cdafb7c8
commit cadd3cdaf9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 1269 additions and 1 deletions

View file

@ -19,6 +19,7 @@ arch_function_generation_params = {
"top_k": 50,
"max_tokens": 512,
"stop_token_ids": [151645],
# "top_logprobs": 10,
}
arch_guard_model_type = {
@ -34,3 +35,4 @@ zero_shot_model = loader.get_zero_shot_model()
prompt_guard_dict = loader.get_prompt_guard(arch_guard_model_type[glb.DEVICE])
arch_guard_handler = ArchGuardHanlder(model_dict=prompt_guard_dict)
# Patterns for function name and parameter parsing

View file

@ -0,0 +1,324 @@
import json
import math
import torch
import random
from typing import Any, Dict, List, Tuple
import itertools
from enum import Enum
# constants
FUNC_NAME_START_PATTERN = ('<tool_call>\n{"name":"', "<tool_call>\n{'name':'")
FUNC_NAME_END_TOKEN = ('",', "',")
TOOL_CALL_TOKEN = "<tool_call>"
FIRST_PARAM_NAME_START_PATTERN = ('"arguments":{"', "'arguments':{'")
PARAMETER_NAME_END_TOKENS = ('":', ':"', "':", ":'")
PARAMETER_NAME_START_PATTERN = (',"', ",'")
PARAMETER_VALUE_START_PATTERN = ('":', "':")
PARAMETER_VALUE_END_TOKEN = ('",', "}}\n", "',")
# Thresholds
class MaskToken(Enum):
FUNCTION_NAME = "f"
PARAMETER_VALUE = "v"
PARAMETER_NAME = "p"
NOT_USED = "e"
TOOL_CALL = "t"
HALLUCINATION_THRESHOLD_DICT = {
MaskToken.TOOL_CALL.value: {"entropy": 0.1, "varentropy": 0.5},
MaskToken.PARAMETER_VALUE.value: {
"entropy": 0.5,
"varentropy": 2.5,
},
}
def check_threshold(entropy: float, varentropy: float, thd: Dict) -> bool:
"""
Check if the given entropy or variance of entropy exceeds the specified thresholds.
Args:
entropy (float): The entropy value to check.
varentropy (float): The variance of entropy value to check.
thd (dict): A dictionary containing the threshold values with keys 'entropy' and 'varentropy'.
Returns:
bool: True if either the entropy or varentropy exceeds their respective thresholds, False otherwise.
"""
return entropy > thd["entropy"] or varentropy > thd["varentropy"]
def calculate_entropy(log_probs: List[float]) -> Tuple[float, float]:
"""
Calculate the entropy and variance of entropy (varentropy) from log probabilities.
Args:
log_probs (list of float): A list of log probabilities.
Returns:
tuple: A tuple containing:
- log_probs (list of float): The input log probabilities as a list.
- entropy (float): The calculated entropy.
- varentropy (float): The calculated variance of entropy.
"""
log_probs = torch.tensor(log_probs)
token_probs = torch.exp(log_probs)
entropy = -torch.sum(log_probs * token_probs, dim=-1) / math.log(2, math.e)
varentropy = torch.sum(
token_probs * (log_probs / math.log(2, math.e)) + entropy.unsqueeze(-1) ** 2,
dim=-1,
)
return entropy.item(), varentropy.item()
def is_parameter_property(
function_description: Dict, parameter_name: str, property_name: str
) -> bool:
"""
Check if a parameter in an API description has a specific property.
Args:
function_description (dict): The API description in JSON format.
parameter_name (str): The name of the parameter to check.
property_name (str): The property to look for (e.g., 'format', 'default').
Returns:
bool: True if the parameter has the specified property, False otherwise.
"""
parameters = function_description.get("properties", {})
parameter_info = parameters.get(parameter_name, {})
return property_name in parameter_info
class HallucinationStateHandler:
"""
A class to handle the state of hallucination detection in token processing.
Attributes:
tokens (list): List of tokens processed.
logprobs (list): List of log probabilities for each token.
state (str): Current state of the handler.
mask (list): List of masks indicating the type of each token.
parameter_name_done (bool): Flag indicating if parameter name extraction is done.
hallucination (bool): Flag indicating if a hallucination is detected.
hallucination_message (str): Message describing the hallucination.
parameter_name (list): List of extracted parameter names.
function_description (dict): Description of functions and their parameters.
token_probs_map (list): List mapping tokens to their entropy and variance of entropy.
"""
def __init__(self, response_iterator=None, function=None):
"""
Initializes the HallucinationStateHandler with default values.
"""
self.tokens: List[str] = []
self.logprobs: List[float] = []
self.state: str = None
self.mask: List[str] = []
self.parameter_name_done: bool = False
self.hallucination: bool = False
self.error_message: str = ""
self.error_type: str = ""
self.parameter_name: List[str] = []
self.token_probs_map: List[Tuple[str, float, float]] = []
self.response_iterator = response_iterator
self._process_function(function)
def _process_function(self, function):
self.function = function
if self.function is None:
raise ValueError("API descriptions not set.")
parameter_names = {}
for func in self.function:
func_name = func["name"]
parameters = func["parameters"]["properties"]
parameter_names[func_name] = list(parameters.keys())
self.function_description = parameter_names
self.function_properties = {x["name"]: x["parameters"] for x in self.function}
def append_and_check_token_hallucination(self, token, logprob):
"""
Check if the given token is hallucinated based on the log probability.
Args:
token (str): The token to check.
logprob (float): The log probability of the token.
Returns:
bool: True if the token is hallucinated, False otherwise.
"""
self.tokens.append(token)
self.logprobs.append(logprob)
self._process_token()
return self.hallucination
def __iter__(self):
return self
def __next__(self):
if self.response_iterator is not None:
try:
r = next(self.response_iterator)
if hasattr(r.choices[0].delta, "content"):
token_content = r.choices[0].delta.content
if token_content:
try:
logprobs = [
p.logprob
for p in r.choices[0].logprobs.content[0].top_logprobs
]
except Exception as e:
raise ValueError(
f"Error extracting logprobs from response: {e}"
)
self.append_and_check_token_hallucination(
token_content, logprobs
)
return token_content
except StopIteration:
raise StopIteration
def _process_token(self):
"""
Processes the current token and updates the state and mask accordingly.
Detects hallucinations based on the token type and log probabilities.
"""
content = "".join(self.tokens).replace(" ", "")
if self.tokens[-1] == TOOL_CALL_TOKEN:
self.mask.append(MaskToken.TOOL_CALL)
self._check_logprob()
# Function name extraction logic
# If the state is function name and the token is not an end token, add to the mask
if self.state == "function_name":
if self.tokens[-1] not in FUNC_NAME_END_TOKEN:
self.mask.append(MaskToken.FUNCTION_NAME)
else:
self.state = None
self._is_function_name_hallucinated()
# Check if the token is a function name start token, change the state
if content.endswith(FUNC_NAME_START_PATTERN):
self.state = "function_name"
# Parameter name extraction logic
# if the state is parameter name and the token is not an end token, add to the mask
if self.state == "parameter_name" and not content.endswith(
PARAMETER_NAME_END_TOKENS
):
self.mask.append(MaskToken.PARAMETER_NAME)
# if the state is parameter name and the token is an end token, change the state, check hallucination and set the flag parameter name done
# The need for parameter name done is to allow the check of parameter value pattern
elif self.state == "parameter_name" and content.endswith(
PARAMETER_NAME_END_TOKENS
):
self.state = None
self._is_parameter_name_hallucinated()
self.parameter_name_done = True
# if the parameter name is done and the token is a parameter name start token, change the state
elif self.parameter_name_done and content.endswith(
PARAMETER_NAME_START_PATTERN
):
self.state = "parameter_name"
# if token is a first parameter value start token, change the state
if content.endswith(FIRST_PARAM_NAME_START_PATTERN):
self.state = "parameter_name"
# Parameter value extraction logic
# if the state is parameter value and the token is not an end token, add to the mask
if self.state == "parameter_value" and not content.endswith(
PARAMETER_VALUE_END_TOKEN
):
# checking if the token is a value token and is not empty
if self.tokens[-1].strip() not in ['"', ""]:
self.mask.append(MaskToken.PARAMETER_VALUE)
# checking if the parameter doesn't have default and the token is the first parameter value token
if (
len(self.mask) > 1
and self.mask[-2] != MaskToken.PARAMETER_VALUE
and not is_parameter_property(
self.function_properties[self.function_name],
self.parameter_name[-1],
"default",
)
):
self._check_logprob()
else:
self.mask.append(MaskToken.NOT_USED)
# if the state is parameter value and the token is an end token, change the state
elif self.state == "parameter_value" and content.endswith(
PARAMETER_VALUE_END_TOKEN
):
self.state = None
# if the parameter name is done and the token is a parameter value start token, change the state
elif self.parameter_name_done and content.endswith(
PARAMETER_VALUE_START_PATTERN
):
self.state = "parameter_value"
# Maintain consistency between stack and mask
# If the mask length is less than tokens, add an not used (e) token to the mask
if len(self.mask) != len(self.tokens):
self.mask.append(MaskToken.NOT_USED)
def _check_logprob(self):
"""
Checks the log probability of the current token and updates the token probability map.
Detects hallucinations based on entropy and variance of entropy.
"""
probs = self.logprobs[-1]
entropy, varentropy = calculate_entropy(probs)
self.token_probs_map.append((self.tokens[-1], entropy, varentropy))
if check_threshold(
entropy, varentropy, HALLUCINATION_THRESHOLD_DICT[self.mask[-1].value]
):
self.hallucination = True
self.error_type = "Hallucination"
self.error_message = (
f"Hallucination: token '{self.tokens[-1]}' is uncertain."
)
def _count_consecutive_token(self, token=MaskToken.PARAMETER_VALUE) -> int:
"""
Counts the number of consecutive occurrences of a given token in the mask.
Args:
token (str): The token to count in the mask.
Returns:
int: The number of consecutive occurrences of the token.
"""
return (
len(list(itertools.takewhile(lambda x: x == token, reversed(self.mask))))
if self.mask and self.mask[-1] == token
else 0
)
def _is_function_name_hallucinated(self):
"""
Checks the extracted function name against the function descriptions.
Detects hallucinations if the function name is not found.
"""
f_len = self._count_consecutive_token(MaskToken.FUNCTION_NAME)
self.function_name = "".join(self.tokens[:-1][-f_len:])
if self.function_name not in self.function_description.keys():
self.error_type = "function_name"
self.error_message = f"Function name '{self.function_name}' not found in given function descriptions."
def _is_parameter_name_hallucinated(self):
"""
Checks the extracted parameter name against the function descriptions.
Detects hallucinations if the parameter name is not found.
"""
p_len = self._count_consecutive_token(MaskToken.PARAMETER_NAME)
parameter_name = "".join(self.tokens[:-1][-p_len:])
self.parameter_name.append(parameter_name)
if parameter_name not in self.function_description[self.function_name]:
self.error_type = "parameter_name"
self.error_message = f"Parameter name '{parameter_name}' not found in given function descriptions."

View file

@ -134,4 +134,4 @@ class ArchFunctionHandler:
fixed_str += opening_bracket[unmatched_opening]
# Attempt to parse the corrected string to ensure its valid JSON
return fixed_str
return fixed_str.replace("'", '"')

View file

@ -0,0 +1,794 @@
[{
"case": "tool_call_halluciation",
"tokens" : ["<tool_call>"],
"expect": 1,
"logprobs": [[-0.3333307206630707,
-1.5310522317886353,
-3.5098977088928223,
-3.9004578590393066,
-5.775152683258057,
-5.814209461212158,
-5.9574151039123535,
-6.0094895362854,
-6.0094895362854,
-6.673445224761963]]
},
{
"case" : "parameter_value_hallucination",
"expect" : 0,
"tokens" : ["<tool_call>",
"\n",
"{'",
"name",
"':",
" '",
"get",
"_current",
"_weather",
"',",
" '",
"arguments",
"':",
" {'",
"location",
"':",
" '",
"Sea",
",",
" Australia",
"',",
" '",
"unit",
"':",
" '",
"c",
"elsius",
"',",
" '",
"days",
"':",
" '",
"1",
"'}}\n",
"</tool_call>"],
"logprobs": [[-0.008103232830762863,
-5.085402488708496,
-6.777836799621582,
-7.558959007263184,
-9.850253105163574,
-10.266852378845215,
-10.540244102478027,
-10.722506523132324,
-10.800618171691895,
-10.917786598205566],
[0.0,
-23.25142478942871,
-25.139137268066406,
-26.2847843170166,
-28.992677688598633,
-29.070789337158203,
-29.55248260498047,
-29.91700553894043,
-30.20341682434082,
-30.307567596435547],
[0.0,
-21.66313934326172,
-23.06916046142578,
-23.32953453063965,
-25.65988540649414,
-25.985353469848633,
-26.519121170043945,
-27.07892417907715,
-27.977216720581055,
-28.458908081054688],
[0.0,
-28.094383239746094,
-28.56305694580078,
-29.109844207763672,
-29.44832992553711,
-31.79170036315918,
-32.0,
-32.05207443237305,
-32.31244659423828,
-32.364524841308594],
[0.0,
-30.489830017089844,
-31.140766143798828,
-31.81774139404297,
-34.525634765625,
-35.8275032043457,
-36.504478454589844,
-39.05614471435547,
-40.123680114746094,
-40.696502685546875],
[0.0,
-25.646865844726562,
-26.66232681274414,
-27.781936645507812,
-28.979660034179688,
-31.140764236450195,
-31.92188835144043,
-31.973962783813477,
-33.04149627685547,
-33.58828353881836],
[0.0,
-23.511798858642578,
-24.136695861816406,
-25.230268478393555,
-25.777053833007812,
-25.80309295654297,
-26.45402717590332,
-26.636289596557617,
-26.740440368652344,
-26.896663665771484],
[0.0,
-22.366153717041016,
-24.683483123779297,
-26.610252380371094,
-26.610252380371094,
-27.313264846801758,
-27.67778778076172,
-28.510986328125,
-28.615135192871094,
-29.13588523864746],
[0.0,
-22.52237319946289,
-24.292919158935547,
-24.344993591308594,
-24.39706802368164,
-24.73555564880371,
-29.943042755126953,
-29.969079971313477,
-30.021154403686523,
-30.0341739654541],
[0.0,
-30.17738151550293,
-30.411718368530273,
-30.88039207458496,
-30.984540939331055,
-31.270952224731445,
-31.895851135253906,
-32.46867370605469,
-32.624900817871094,
-33.484134674072266],
[0.0,
-28.146459579467773,
-29.396255493164062,
-30.099267959594727,
-31.127744674682617,
-31.179821014404297,
-32.807159423828125,
-33.7445068359375,
-33.770545959472656,
-34.069976806640625],
[0.0,
-26.323841094970703,
-26.558177947998047,
-30.515867233276367,
-30.932466506958008,
-31.37510108947754,
-31.531326293945312,
-31.70056915283203,
-32.065093994140625,
-32.364524841308594],
[0.0,
-26.922698974609375,
-30.28152847290039,
-31.505287170410156,
-33.30187225341797,
-33.73148727416992,
-34.27827453613281,
-34.33034896850586,
-34.460533142089844,
-34.720909118652344],
[0.0,
-21.532955169677734,
-26.94873809814453,
-29.109848022460938,
-30.80228042602539,
-31.55736541748047,
-33.484134674072266,
-34.681854248046875,
-35.384864807128906,
-35.853538513183594],
[0.0,
-19.502033233642578,
-20.46541976928711,
-24.110658645629883,
-24.501218795776367,
-25.256305694580078,
-25.82912826538086,
-25.881202697753906,
-26.063465118408203,
-26.063465118408203],
[0.0,
-24.37103271484375,
-25.256305694580078,
-25.933277130126953,
-26.714401245117188,
-28.2506103515625,
-31.010576248168945,
-32.07810974121094,
-34.62977981567383,
-35.241661071777344],
[-1.1920922133867862e-06,
-14.398697853088379,
-14.424736976623535,
-17.158666610717773,
-17.41904067993164,
-18.200162887573242,
-18.434499740600586,
-18.66883659362793,
-19.71033477783203,
-19.71033477783203],
[-0.0001445904199499637,
-8.98305892944336,
-11.35246467590332,
-13.1490478515625,
-13.669795989990234,
-14.073375701904297,
-14.516012191772461,
-14.555068969726562,
-15.622602462768555,
-15.635622024536133],
[-0.44747352600097656,
-1.0202960968017578,
-8.467000961303711,
-10.914518356323242,
-11.25300407409668,
-11.435266494750977,
-12.346576690673828,
-13.075624465942383,
-13.12769889831543,
-13.231849670410156],
[-3.123767137527466,
-1.1188862323760986,
-1.639634370803833,
-2.0562336444854736,
-2.8633930683135986,
-2.9675419330596924,
-3.4882919788360596,
-3.69659161567688,
-4.217339515686035,
-4.243376731872559],
[-7.199982064776123e-05,
-9.76410961151123,
-11.144091606140137,
-16.507802963256836,
-17.132701873779297,
-17.44515037536621,
-17.9138240814209,
-18.33042335510254,
-18.9162654876709,
-19.39795684814453],
[0.0,
-22.991050720214844,
-23.824249267578125,
-24.969894409179688,
-25.46460723876953,
-25.829130172729492,
-26.480066299438477,
-26.909683227539062,
-27.33930206298828,
-27.391376495361328],
[-0.21928852796554565,
-1.625309705734253,
-9.775025367736816,
-12.977627754211426,
-16.388530731201172,
-17.091541290283203,
-19.044347763061523,
-19.38283348083496,
-19.460947036743164,
-19.59113311767578],
[0.0,
-24.006507873535156,
-27.443450927734375,
-27.729862213134766,
-28.12042236328125,
-28.276647567749023,
-28.927583694458008,
-30.099267959594727,
-31.479251861572266,
-32.07810974121094],
[0.0,
-18.17412567138672,
-18.772987365722656,
-21.689178466796875,
-21.92351531982422,
-23.7200984954834,
-23.79821014404297,
-23.79821014404297,
-24.032546997070312,
-25.308382034301758],
[-0.12947827577590942,
-2.1083219051361084,
-12.419143676757812,
-15.23118782043457,
-15.595710754394531,
-15.830047607421875,
-17.001731872558594,
-17.60059356689453,
-18.121341705322266,
-18.251529693603516],
[0.0,
-19.449962615966797,
-24.371034622192383,
-24.917821884155273,
-25.529701232910156,
-25.85516929626465,
-26.037429809570312,
-26.115543365478516,
-26.623271942138672,
-26.649309158325195],
[-0.03332124650478363,
-3.4181859493255615,
-15.759925842285156,
-15.812002182006836,
-16.593124389648438,
-17.894996643066406,
-18.09027671813965,
-18.79328727722168,
-19.144792556762695,
-20.147233963012695],
[0.0,
-21.142393112182617,
-22.157852172851562,
-23.511798858642578,
-24.657445907592773,
-25.021968841552734,
-25.5427188873291,
-25.59479331970215,
-25.75101661682129,
-25.95931625366211],
[0.0,
-23.04312515258789,
-24.94385528564453,
-26.323841094970703,
-27.54759979248047,
-28.563060760498047,
-29.786819458007812,
-30.620018005371094,
-30.69812774658203,
-31.08869171142578],
[0.0,
-26.167617797851562,
-28.771360397338867,
-29.55248260498047,
-30.906429290771484,
-31.114728927612305,
-31.414159774780273,
-31.622459411621094,
-31.713590621948242,
-31.726608276367188],
[-0.05012698099017143,
-3.018392562866211,
-11.740934371948242,
-13.146955490112305,
-13.797887802124023,
-14.943536758422852,
-16.037107467651367,
-16.375595092773438,
-16.714080810546875,
-17.36501693725586],
[-0.9704352021217346,
-0.7360983490943909,
-2.1941938400268555,
-4.225115776062012,
-5.0062360763549805,
-5.2666120529174805,
-5.839434623718262,
-7.2714948654174805,
-8.33902645111084,
-8.495253562927246],
[-0.014467108063399792,
-4.258565902709961,
-8.789079666137695,
-10.429437637329102,
-10.793962478637695,
-11.835458755493164,
-11.939607620239258,
-13.31959342956543,
-13.866378784179688,
-15.038063049316406],
[0.0,
-20.08787727355957,
-21.350692749023438,
-21.415786743164062,
-21.50691795349121,
-21.50691795349121,
-22.7176570892334,
-24.13669776916504,
-24.188772201538086,
-24.34499740600586]]
},
{
"case": "fail_case",
"expect" : 0,
"tokens" : ["<tool_call>",
"\n",
"{'",
"name",
"':",
" '",
"get",
"_current",
"_weather",
"',",
" '",
"arguments",
"':",
" {'",
"location",
"':",
" '",
"Seattle",
",",
" WA",
"',",
" '",
"unit",
"':",
" '",
"c",
"elsius",
"',",
" '",
"days",
"':",
" '",
"7",
"'}}\n",
"</tool_call>"],
"logprobs":[[-0.00013815402053296566,
-9.113236427307129,
-10.571331977844238,
-14.099404335021973,
-14.28166675567627,
-15.583537101745605,
-15.81787395477295,
-16.143341064453125,
-16.143341064453125,
-16.260509490966797],
[0.0,
-26.896663665771484,
-27.32628059387207,
-27.41741180419922,
-32.07810974121094,
-32.07810974121094,
-32.28641128540039,
-32.29943084716797,
-32.44263458251953,
-32.520748138427734],
[0.0,
-22.444263458251953,
-24.527257919311523,
-27.15703773498535,
-28.016273498535156,
-28.2506103515625,
-28.693246841430664,
-29.070789337158203,
-29.565500259399414,
-29.812854766845703],
[0.0,
-27.860050201416016,
-28.641170501708984,
-29.448333740234375,
-30.932466506958008,
-31.63547706604004,
-32.33848571777344,
-32.85923767089844,
-33.17168426513672,
-33.45809555053711],
[0.0,
-31.81774139404297,
-31.895854949951172,
-32.05207824707031,
-35.43694305419922,
-36.3482551574707,
-38.61351013183594,
-39.26444625854492,
-40.61839294433594,
-41.71196365356445],
[0.0,
-27.33930206298828,
-27.834014892578125,
-28.849472045898438,
-30.567943572998047,
-32.98942565917969,
-33.067535400390625,
-33.067535400390625,
-35.67127990722656,
-35.69731903076172],
[0.0,
-25.33441925048828,
-26.063465118408203,
-26.219690322875977,
-26.2457275390625,
-26.53213882446289,
-27.365337371826172,
-28.354759216308594,
-28.667207717895508,
-28.74532127380371],
[0.0,
-24.423107147216797,
-24.579330444335938,
-26.81855010986328,
-28.12042236328125,
-28.32872200012207,
-28.61513328552246,
-29.16191864013672,
-29.187957763671875,
-29.240032196044922],
[0.0,
-22.027664184570312,
-23.850284576416016,
-23.980472564697266,
-24.292922973632812,
-24.787633895874023,
-29.279088973999023,
-29.55248260498047,
-29.903987884521484,
-30.190399169921875],
[0.0,
-31.609439849853516,
-31.817739486694336,
-32.54678726196289,
-32.676971435546875,
-32.781124114990234,
-32.98942565917969,
-33.106590270996094,
-33.57526397705078,
-34.369407653808594],
[0.0,
-29.34418296813965,
-29.63059425354004,
-30.021156311035156,
-30.984540939331055,
-33.21073913574219,
-34.30431365966797,
-34.56468963623047,
-34.70789337158203,
-34.79902648925781],
[0.0,
-25.438566207885742,
-25.69894027709961,
-30.190397262573242,
-30.802276611328125,
-31.58340072631836,
-31.609437942504883,
-31.64849281311035,
-31.973960876464844,
-32.29943084716797],
[0.0,
-27.157039642333984,
-32.104148864746094,
-32.33848571777344,
-34.04393768310547,
-34.12205505371094,
-34.40846252441406,
-34.42148208618164,
-34.772987365722656,
-34.87713623046875],
[0.0,
-24.813671112060547,
-26.974777221679688,
-31.010578155517578,
-31.08869171142578,
-32.1822624206543,
-35.33279037475586,
-35.489013671875,
-36.999183654785156,
-37.88446044921875],
[0.0,
-20.46541976928711,
-20.647682189941406,
-23.069164276123047,
-24.136699676513672,
-25.438570022583008,
-25.646869659423828,
-26.193655014038086,
-26.297805786132812,
-26.506103515625],
[0.0,
-27.18307113647461,
-28.30268096923828,
-28.56305694580078,
-29.526439666748047,
-32.416595458984375,
-35.202598571777344,
-36.426361083984375,
-39.31651306152344,
-39.38160705566406],
[0.0,
-18.7469482421875,
-20.100894927978516,
-21.402767181396484,
-21.428804397583008,
-22.20992660522461,
-22.34011459350586,
-22.730674743652344,
-23.069162368774414,
-23.980472564697266],
[-3.576278118089249e-07,
-15.2579345703125,
-16.481693267822266,
-17.991863250732422,
-19.215621948242188,
-20.25712013244629,
-21.350692749023438,
-22.314077377319336,
-22.496337890625,
-22.938974380493164],
[-0.08506780862808228,
-2.506549835205078,
-14.848289489746094,
-15.473188400268555,
-16.33242416381836,
-16.358461380004883,
-16.566761016845703,
-17.03543472290039,
-17.686370849609375,
-17.816556930541992],
[-0.0194891095161438,
-4.445854187011719,
-5.591499328613281,
-5.956024169921875,
-6.685070037841797,
-13.142353057861328,
-13.558952331542969,
-15.173273086547852,
-15.303461074829102,
-15.85024642944336],
[-0.0005990855861455202,
-7.4212646484375,
-15.675132751464844,
-15.72720718383789,
-16.76870346069336,
-16.76870346069336,
-17.706050872802734,
-18.669435501098633,
-19.398483276367188,
-19.658857345581055],
[0.0,
-24.110658645629883,
-25.829130172729492,
-26.011390686035156,
-26.011390686035156,
-26.532140731811523,
-26.58421516418457,
-27.651750564575195,
-27.75589942932129,
-28.055330276489258],
[-1.1408883333206177,
-0.38580334186553955,
-7.494022369384766,
-12.519245147705078,
-14.576202392578125,
-16.034297943115234,
-16.945608139038086,
-17.908992767333984,
-18.664077758789062,
-19.34105110168457],
[0.0,
-26.688365936279297,
-29.83889389038086,
-30.177383422851562,
-30.64605712890625,
-31.244916915893555,
-31.270954132080078,
-32.83319854736328,
-34.655818939208984,
-34.89015579223633],
[0.0,
-18.929210662841797,
-19.16354751586914,
-23.589908599853516,
-24.683481216430664,
-24.995929718017578,
-25.516677856445312,
-25.542715072631836,
-25.77705192565918,
-26.063465118408203],
[-0.2519786059856415,
-1.5017764568328857,
-12.437495231628418,
-15.457839012145996,
-15.744250297546387,
-16.837820053100586,
-17.41064453125,
-17.56686782836914,
-17.61894416809082,
-18.035541534423828],
[0.0,
-20.517494201660156,
-24.683483123779297,
-25.67290496826172,
-26.58421516418457,
-27.651750564575195,
-27.781936645507812,
-27.912124633789062,
-28.09438705444336,
-28.445892333984375],
[-3.40932747349143e-05,
-10.284820556640625,
-18.252273559570312,
-20.17904281616211,
-21.663175582885742,
-22.027700424194336,
-22.288074493408203,
-22.704673767089844,
-23.12127113342285,
-23.277496337890625],
[0.0,
-22.60049057006836,
-25.46460723876953,
-25.829130172729492,
-26.063467025756836,
-27.287227630615234,
-27.391376495361328,
-27.4694881439209,
-27.67778778076172,
-28.055330276489258],
[0.0,
-23.902362823486328,
-28.823436737060547,
-29.240036010742188,
-29.31814956665039,
-29.917007446289062,
-30.021160125732422,
-31.21887969970703,
-32.416603088378906,
-32.416603088378906],
[0.0,
-28.641170501708984,
-31.947925567626953,
-32.59886169433594,
-33.848655700683594,
-34.109031677246094,
-34.73393249511719,
-35.02033996582031,
-35.02033996582031,
-36.074859619140625],
[-0.013183215633034706,
-4.335395336151123,
-19.619365692138672,
-20.035964965820312,
-20.244266510009766,
-21.311800003051758,
-21.441987991333008,
-22.561595916748047,
-23.108383178710938,
-23.264606475830078],
[-8.344646857949556e-07,
-14.190400123596191,
-15.9088716506958,
-18.17412567138672,
-18.46053695678711,
-18.46053695678711,
-18.512611389160156,
-18.90317153930664,
-19.059398651123047,
-19.085433959960938],
[0.0,
-17.70545196533203,
-18.903175354003906,
-20.829944610595703,
-22.574451446533203,
-22.860862731933594,
-23.069162368774414,
-23.32953643798828,
-23.694061279296875,
-24.188772201538086],
[0.0,
-20.022781372070312,
-21.038240432739258,
-21.220502853393555,
-22.496337890625,
-22.769729614257812,
-23.589908599853516,
-23.65500259399414,
-23.94141387939453,
-24.266881942749023]]
}
]

View file

@ -0,0 +1,148 @@
import json
from app.function_calling.hallucination_handler import HallucinationStateHandler
import pytest
import os
# Get the directory of the current file
current_dir = os.path.dirname(__file__)
# Construct the full path to the JSON file
json_file_path = os.path.join(current_dir, "test_cases.json")
with open(json_file_path) as f:
test_cases = json.load(f)
get_weather_api = {
"type": "function",
"function": {
"name": "get_current_weather",
"description": "Get current weather at a location.",
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "str",
"description": "The location to get the weather for",
"format": "City, State",
},
"unit": {
"type": "str",
"description": "The unit to return the weather in.",
"enum": ["celsius", "fahrenheit"],
"default": "celsius",
},
"days": {
"type": "str",
"description": "the number of days for the request.",
},
},
"required": ["location", "days"],
},
},
}
function_description = get_weather_api["function"]
if type(function_description) != list:
function_description = [get_weather_api["function"]]
@pytest.mark.parametrize("case", test_cases)
def test_hallucination(case):
state = HallucinationStateHandler(
response_iterator=None, function=function_description
)
for token, logprob in zip(case["tokens"], case["logprobs"]):
if token != "</tool_call>":
state.append_and_check_token_hallucination(token, logprob)
if state.hallucination:
break
assert state.hallucination == case["expect"]
@pytest.mark.parametrize("is_hallucinate_sample", [True, False])
def test_hallucination_prompt(is_hallucinate_sample):
TASK_PROMPT = """
You are a helpful assistant.
""".strip()
TOOL_PROMPT = """
# Tools
You may call one or more functions to assist with the user query.
You are provided with function signatures within <tools></tools> XML tags:
<tools>
{tool_text}
</tools>
""".strip()
FORMAT_PROMPT = """
For each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:
<tool_call>
{"name": <function-name>, "arguments": <args-json-object>}
</tool_call>
""".strip()
def convert_tools(tools):
return "\n".join([json.dumps(tool) for tool in tools])
def format_prompt(tools):
tool_text = convert_tools(tools)
return (
TASK_PROMPT
+ "\n\n"
+ TOOL_PROMPT.format(tool_text=tool_text)
+ "\n\n"
+ FORMAT_PROMPT
+ "\n"
)
openai_format_tools = [get_weather_api]
system_prompt = format_prompt(openai_format_tools)
from openai import OpenAI
client = OpenAI(base_url="https://api.fc.archgw.com/v1", api_key="EMPTY")
# List models API
model = client.models.list().data[0].id
assert model == "Arch-Function"
if not is_hallucinate_sample:
messages = [
{"role": "system", "content": system_prompt},
# {"role": "user", "content": "can you help me check weather?"},
{"role": "user", "content": "How is the weather in Seattle in 7 days?"},
# {"role": "assistant", "content": "Of course!"},
# {"role": "user", "content": "Seattle please"}
]
else:
messages = [
{"role": "system", "content": system_prompt},
# {"role": "user", "content": "can you help me check weather?"},
{"role": "user", "content": "How is the weather in Seattle in days?"},
# {"role": "assistant", "content": "Of course!"},
# {"role": "user", "content": "Seattle please"}
]
extra_body = {
"temperature": 0.6,
"top_p": 1.0,
"top_k": 50,
# "continue_final_message": True,
# "add_generation_prompt": False,
"logprobs": True,
"top_logprobs": 10,
}
resp = client.chat.completions.create(
model="Arch-Function", messages=messages, extra_body=extra_body, stream=True
)
hallu = HallucinationStateHandler(
response_iterator=resp, function=function_description
)
for token in hallu:
assert len(hallu.tokens) >= 0
assert hallu.hallucination == is_hallucinate_sample