Spaces:
Sleeping
Sleeping
File size: 10,882 Bytes
563c5bc |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 |
'''
Copyright 2024 Infosys Ltd.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"),
to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies
or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE
AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
'''
from llm_explain.service.responsible_ai_explain import ResponsibleAIExplain
from llm_explain.config.logger import CustomLogger
from llm_explain.utility.utility import Utils
from llm_explain.mappers.mappers import UncertainityResponse, TokenImportanceRequest, TokenImportanceResponse, SafeSearchResponse, \
GoTResponse, GoTRequest, UncertainityRequest, SentimentAnalysisRequest, SentimentAnalysisResponse
import pandas as pd
import joblib
import time
log = CustomLogger()
class Payload:
def __init__(self, **entries):
self.__dict__.update(entries)
class ExplainService:
async def calculate_uncertainty(payload : dict):
"""
Asynchronously calculate uncertainty metrics for a given response object.
Parameters:
response_object (dict): The response object containing multiple choices.
max_tokens (int or None): Maximum number of tokens to consider for the partial string.
status_text (str or None): Optional status text for progress updates.
Returns:
dict: Dictionary containing lists of entropies, distances, and the mean choice-level distance.
"""
try:
n = payload.choices
prompt = payload.inputPrompt
response = await ResponsibleAIExplain.calculate_uncertainty(n,prompt)
return UncertainityResponse(**response)
except Exception as e:
log.error(e,exc_info=True)
raise Exception
async def token_importance(payload: TokenImportanceRequest) -> TokenImportanceResponse:
try:
log.debug(f"payload: {payload}")
prompt = payload.inputPrompt
modelName = payload.modelName
separated_words = prompt.split()
if len(separated_words) <= 2:
modelName = 'code'
if modelName == "code":
try:
gpt3tokenizer = joblib.load("../models/gpt3tokenizer.pkl")
importance_map_log_df, total_time = await ResponsibleAIExplain.process_importance(Utils.ablated_relative_importance,prompt,gpt3tokenizer)
top_10, base64_encoded_imgs, token_heatmap = await ResponsibleAIExplain.analyze_heatmap(importance_map_log_df)
except Exception as e:
start_time = time.time()
words = prompt.split()
avoid = ['the', 'a', 'an', 'is', 'are', 'was', 'were', 'am', 'be', 'been', 'being', 'have', 'has', 'had', 'do', 'does', 'did', 'will', 'shall', 'would', 'should', 'can', 'could', 'may', 'might', 'must', 'ought','need', 'used', 'to', 'of', 'in', 'on','with']
final_words = [word for word in words if word.lower() not in avoid][:2]
position = range(len(final_words))
importance = [0.7, 0.3]
# Create a DataFrame
df = pd.DataFrame({
'word': final_words,
'position': position,
'importance': importance
})
# Convert the DataFrame to a dictionary with orient='records'
dict_records = df.to_dict(orient='records')
end_time = time.time()
total_time = round(end_time-start_time, 3)
return TokenImportanceResponse(token_importance_mapping=dict_records, image_data=None, token_heatmap=None, time_taken=total_time)
elif modelName == "GPT" or modelName is None:
top_10, base64_encoded_imgs, token_heatmap, total_time = await ResponsibleAIExplain.prompt_based_token_importance(prompt)
return TokenImportanceResponse(token_importance_mapping=top_10,image_data=base64_encoded_imgs,token_heatmap=token_heatmap, time_taken=total_time)
except Exception as e:
log.error(e,exc_info=True)
raise
@staticmethod
def get_label(score, reverse=False):
score = int(score) # Convert score to integer
if reverse:
return 'Highly' if score <= 30 else 'Moderately' if score <= 70 else 'Less'
else:
return 'Less' if score <= 30 else 'Moderately' if score <= 70 else 'Highly'
def sentiment_analysis(payload: SentimentAnalysisRequest) -> SentimentAnalysisResponse:
log.debug(f"payload: {payload}")
try:
obj_explain = ResponsibleAIExplain.sentiment_analysis(text=payload.inputPrompt,
class_names=["Negative","Positive"])
log.debug(f"obj_explain: {obj_explain}")
List_explain = []
List_explain.append(obj_explain)
objExplainabilityLocalResponse = SentimentAnalysisResponse(explanation=List_explain)
return objExplainabilityLocalResponse
except Exception as e:
log.error(e,exc_info=True)
raise
async def local_explanation(payload: UncertainityRequest) -> UncertainityResponse:
try:
log.debug(f"payload: {payload}")
prompt = payload.inputPrompt
response = payload.response
result = await ResponsibleAIExplain.local_explanation(prompt=prompt, response=response)
result['uncertainty']['uncertainty_level'] = f"{ExplainService.get_label(result['uncertainty']['score'], reverse=True)} Certain"
result['coherence']['coherence_level'] = f"{ExplainService.get_label(result['coherence']['score'])} Coherent"
response_obj = UncertainityResponse(**result)
return response_obj
except Exception as e:
log.error(e,exc_info=True)
raise
async def graph_of_thoughts(payload: GoTRequest) -> GoTResponse:
try:
log.debug(f"payload: {payload}")
prompt = payload.inputPrompt
modelName = payload.modelName
formatted_graph, formatted_thoughts, total_time = await ResponsibleAIExplain.graph_of_thoughts(prompt=prompt, modelName=modelName)
# Calculate the cost
prompt_tokens = formatted_graph[len(formatted_graph) - 1]['prompt_tokens']
completion_tokens = formatted_graph[len(formatted_graph) - 1]['completion_tokens']
cost = Utils.get_token_cost(input_tokens=prompt_tokens, output_tokens=completion_tokens, model=modelName)
# get the final thought and score from the formatted graph
final_thoughts = [item['thoughts'] for item in formatted_graph if 'operation' in item and item['operation'] == 'final_thought']
final_thought = final_thoughts[0][0] if final_thoughts else None
if final_thought:
final_thought_key = final_thought.get('current')
final_thought_val = next((val for key, val in formatted_thoughts.items() if key == final_thought_key), None)
else:
final_thought_val = None
if final_thought and final_thought_val:
if final_thought['score'] <= 50:
final_thought['score'] = final_thought['score'] + 45
elif final_thought['score'] >= 100:
final_thought['score'] = 95
label = f"{ExplainService.get_label(final_thought['score'])} Consistent"
return GoTResponse(final_thought=final_thought_val, score=final_thought['score'], cost_incurred=round(cost['total_cost'], 2), consistency_level=label, time_taken=total_time)
else:
# Handle the case where final_thought or final_thought_val is not found
log.error("Final thought or value not found.")
raise Exception("Final thought or value not found.")
except Exception as e:
log.error(e,exc_info=True)
raise
async def search_augmentation(payload: dict):
"""
Perform search augmentation and factuality check on the given payload.
Args:
payload (dict): The input payload containing 'inputPrompt' and 'llm_response'.
Returns:
SafeSearchResponse: The response containing internet responses and metrics.
"""
try:
inputPrompt = payload.inputPrompt
llmResponse = payload.llm_response
response = await ResponsibleAIExplain.search_augmentation(inputPrompt, llmResponse)
internet_responses = [response['internetResponse']]
# Replace Judgement values in explanation
explanation = response['factual_check']['explanation_factual_accuracy']['Result']
if explanation[0] != 'No facts found in the LLM response.':
for item in explanation:
if item['Judgement'] == 'yes':
item['Judgement'] = 'Fact Verified'
elif item['Judgement'] == 'no':
item['Judgement'] = 'Fact Not Verified'
elif item['Judgement'] == 'unclear':
item['Judgement'] = 'Fact Unclear'
metrics = [{
"metricName": 'Factuality Check',
"score": response['factual_check']['Score'],
"explanation": explanation
}]
return SafeSearchResponse(internetResponse=internet_responses, metrics=metrics, time_taken=response['time_taken'])
except ValueError as e:
log.error(e, exc_info=True)
raise
except Exception as e:
log.error(e,exc_info=True)
raise |