Spaces:
Running
Running
File size: 11,825 Bytes
567930d d0291ae 806953a e91ac58 d174ce3 806953a d0291ae 806953a e91ac58 806953a b8abf64 567930d e91ac58 ae215ea e91ac58 d0291ae e91ac58 567930d ae215ea e91ac58 567930d e91ac58 567930d e91ac58 567930d d174ce3 e91ac58 567930d e91ac58 567930d e91ac58 567930d e91ac58 d0291ae e91ac58 567930d d0291ae e91ac58 d0291ae e91ac58 d174ce3 e91ac58 567930d e91ac58 d174ce3 e91ac58 9d06861 e91ac58 567930d e91ac58 d0291ae e91ac58 d0291ae 567930d e91ac58 9d06861 567930d ae215ea 9d06861 ae215ea 9d06861 e91ac58 9d06861 e91ac58 567930d 9d06861 e91ac58 567930d e91ac58 ae215ea 9d06861 e91ac58 567930d |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 |
import os, time, json, typing
from dataclasses import dataclass
# import vertexai
from vertexai.language_models import TextGenerationModel
from vertexai.generative_models._generative_models import HarmCategory, HarmBlockThreshold
from vertexai.language_models import TextGenerationModel
# from vertexai.preview.generative_models import GenerativeModel
from langchain.output_parsers.retry import RetryWithErrorOutputParser
# from langchain.schema import HumanMessage
from langchain.prompts import PromptTemplate
from langchain_core.output_parsers import JsonOutputParser
# from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_google_vertexai import VertexAI
from langchain_core.messages import BaseMessage, HumanMessage
from vouchervision.utils_LLM import SystemLoadMonitor, run_tools, count_tokens, save_individual_prompt, sanitize_prompt
from vouchervision.utils_LLM_JSON_validation import validate_and_align_JSON_keys_with_template
#https://cloud.google.com/vertex-ai/docs/python-sdk/use-vertex-ai-python-sdk
#pip install --upgrade google-cloud-aiplatform
# from google.cloud import aiplatform
#### have to authenticate gcloud
# gcloud auth login
# gcloud config set project XXXXXXXXX
# https://cloud.google.com/docs/authentication
from pydantic import BaseModel
from langchain_core.prompt_values import PromptValue as BasePromptValue
class PromptValueWrapper(BaseModel):
prompt_str: str
def to_string(self) -> str:
return self.prompt_str
def to_messages(self):
return [HumanMessage(content=self.prompt_str)]
class GooglePalm2Handler:
RETRY_DELAY = 10 # Wait 10 seconds before retrying
MAX_RETRIES = 3 # Maximum number of retries
TOKENIZER_NAME = 'gpt-4'
VENDOR = 'google'
STARTING_TEMP = 0.5
def __init__(self, cfg, logger, model_name, JSON_dict_structure, config_vals_for_permutation):
self.cfg = cfg
self.tool_WFO = self.cfg['leafmachine']['project']['tool_WFO']
self.tool_GEO = self.cfg['leafmachine']['project']['tool_GEO']
self.tool_wikipedia = self.cfg['leafmachine']['project']['tool_wikipedia']
self.logger = logger
self.model_name = model_name
self.JSON_dict_structure = JSON_dict_structure
self.config_vals_for_permutation = config_vals_for_permutation
self.monitor = SystemLoadMonitor(logger)
self.parser = JsonOutputParser()
# Define the prompt template
self.prompt = PromptTemplate(
template="Answer the user query.\n{format_instructions}\n{query}\n",
input_variables=["query"],
partial_variables={"format_instructions": self.parser.get_format_instructions()},
)
self._set_config()
def _set_config(self):
# vertexai.init(project=os.environ['PALM_PROJECT_ID'], location=os.environ['PALM_LOCATION'])
if self.config_vals_for_permutation:
self.starting_temp = float(self.config_vals_for_permutation.get('google').get('temperature'))
self.config = {
'max_output_tokens': self.config_vals_for_permutation.get('google').get('max_output_tokens'),
'temperature': self.starting_temp,
'top_k': self.config_vals_for_permutation.get('google').get('top_k'),
'top_p': self.config_vals_for_permutation.get('google').get('top_p'),
}
else:
self.starting_temp = float(self.STARTING_TEMP)
self.config = {
"max_output_tokens": 1024,
"temperature": self.starting_temp,
"top_k": 1,
"top_p": 1.0,
}
self.temp_increment = float(0.2)
self.adjust_temp = self.starting_temp
self.safety_settings = {
HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_NONE,
HarmCategory.HARM_CATEGORY_HATE_SPEECH: HarmBlockThreshold.BLOCK_NONE,
HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_NONE,
HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: HarmBlockThreshold.BLOCK_NONE,
}
self._build_model_chain_parser()
def _adjust_config(self):
new_temp = self.adjust_temp + self.temp_increment
if self.json_report:
self.json_report.set_text(text_main=f'Incrementing temperature from {self.adjust_temp} to {new_temp}')
self.logger.info(f'Incrementing temperature from {self.adjust_temp} to {new_temp}')
self.adjust_temp += self.temp_increment
self.config['temperature'] = self.adjust_temp
def _reset_config(self):
if self.json_report:
self.json_report.set_text(text_main=f'Resetting temperature from {self.adjust_temp} to {self.starting_temp}')
self.logger.info(f'Resetting temperature from {self.adjust_temp} to {self.starting_temp}')
self.adjust_temp = self.starting_temp
self.config['temperature'] = self.starting_temp
# def _build_model_chain_parser(self):
# # Instantiate the parser and the retry parser
# # self.llm_model = ChatGoogleGenerativeAI(model=self.model_name)
# self.llm_model = VertexAI(model=self.model_name,
# max_output_tokens=self.config.get('max_output_tokens'),
# temperature=self.config.get('temperature'),
# top_k=self.config.get('top_k'),
# top_p=self.config.get('top_p'))
# self.retry_parser = RetryWithErrorOutputParser.from_llm(
# parser=self.parser,
# llm=self.llm_model,
# max_retries=self.MAX_RETRIES)
# # Prepare the chain
# self.chain = self.prompt | self.call_google_palm2
def _build_model_chain_parser(self):
# Instantiate the parser and the retry parser
self.llm_model = VertexAI(model=self.model_name,
max_output_tokens=self.config.get('max_output_tokens'),
temperature=self.config.get('temperature'),
top_k=self.config.get('top_k'),
top_p=self.config.get('top_p'))
self.retry_parser = RetryWithErrorOutputParser.from_llm(
llm=self.llm_model,
parser=self.parser,
max_retries=self.MAX_RETRIES
)
# Prepare the chain
self.chain = self.prompt | self.call_google_palm2
# Define a function to format the input for Google PaLM call
# https://cloud.google.com/vertex-ai/docs/generative-ai/migrate/migrate-palm-to-gemini?_ga=2.225326234.-1652490527.1705461451&_gac=1.186295771.1706291573.CjwKCAiAzc2tBhA6EiwArv-i6QCpx7xTP0yrBy9KKSwno3QXOWUe14mbp9RGZO0ShcbtFqyXii2PnRoCywgQAvD_BwE
def call_google_palm2(self, prompt_text):
model = TextGenerationModel.from_pretrained(self.model_name)
response = model.predict(prompt_text.text,
max_output_tokens=self.config.get('max_output_tokens'),
temperature=self.config.get('temperature'),
top_k=self.config.get('top_k'),
top_p=self.config.get('top_p'))
# model = GenerativeModel(self.model_name)
# response = model.generate_content(prompt_text.text,generation_config=self.config, safety_settings=self.safety_settings, stream=False)
return response.text
def call_llm_api_GooglePalm2(self, prompt_template, json_report, paths):
_____, ____, _, __, ___, json_file_path_wiki, txt_file_path_ind_prompt = paths
self.json_report = json_report
if json_report:
self.json_report.set_text(text_main=f'Sending request to {self.model_name}')
self.monitor.start_monitoring_usage()
nt_in = 0
nt_out = 0
ind = 0
while ind < self.MAX_RETRIES:
ind += 1
try:
model_kwargs = {"temperature": self.adjust_temp}
# Invoke the chain to generate prompt text
response = self.chain.invoke({"query": prompt_template, "model_kwargs": model_kwargs})
# Clean up the response by removing any Markdown formatting
response_text = response.strip('```JSON\n').strip('\n```')
output = json.loads(response_text)
# # Use retry_parser to parse the response with retry logic
# try:
# output = self.retry_parser.parse_with_prompt(response, prompt_value=PromptValue(prompt_template))
# except:
# try:
# output = self.retry_parser.parse_with_prompt(response, prompt_value=prompt_template)
# except:
# try:
# output = json.loads(response)
# except Exception as e:
# print(e)
# output = None
if output is None:
self.logger.error(f'[Attempt {ind}] Failed to extract JSON from:\n{response}')
self._adjust_config()
else:
nt_in = count_tokens(prompt_template, self.VENDOR, self.TOKENIZER_NAME)
nt_out = count_tokens(response, self.VENDOR, self.TOKENIZER_NAME)
output = validate_and_align_JSON_keys_with_template(output, self.JSON_dict_structure)
if output is None:
self.logger.error(f'[Attempt {ind}] Failed to extract JSON from:\n{response}')
self._adjust_config()
else:
self.monitor.stop_inference_timer() # Starts tool timer too
if self.json_report:
self.json_report.set_text(text_main=f'Working on WFO, Geolocation, Links')
output_WFO, WFO_record, output_GEO, GEO_record = run_tools(output, self.tool_WFO, self.tool_GEO, self.tool_wikipedia, json_file_path_wiki)
save_individual_prompt(sanitize_prompt(prompt_template), txt_file_path_ind_prompt)
self.logger.info(f"Formatted JSON:\n{json.dumps(output,indent=4)}")
usage_report = self.monitor.stop_monitoring_report_usage()
if self.adjust_temp != self.starting_temp:
self._reset_config()
if self.json_report:
self.json_report.set_text(text_main=f'LLM call successful')
return output, nt_in, nt_out, WFO_record, GEO_record, usage_report
except Exception as e:
self.logger.error(f'{e}')
self._adjust_config()
time.sleep(self.RETRY_DELAY)
self.logger.info(f"Failed to extract valid JSON after [{ind}] attempts")
if self.json_report:
self.json_report.set_text(text_main=f'Failed to extract valid JSON after [{ind}] attempts')
self.monitor.stop_inference_timer() # Starts tool timer too
usage_report = self.monitor.stop_monitoring_report_usage()
self._reset_config()
if self.json_report:
self.json_report.set_text(text_main=f'LLM call failed')
return None, nt_in, nt_out, None, None, usage_report
|