Spaces:
Running
Running
File size: 8,671 Bytes
9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 e91ac58 9d06861 |
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 |
import time, torch, json
from langchain.prompts import PromptTemplate
from langchain_openai import ChatOpenAI, OpenAI
from langchain.schema import HumanMessage
from langchain_core.output_parsers import JsonOutputParser
from langchain.output_parsers import RetryWithErrorOutputParser
from vouchervision.utils_LLM import SystemLoadMonitor, count_tokens, save_individual_prompt
from vouchervision.utils_LLM_JSON_validation import validate_and_align_JSON_keys_with_template
from vouchervision.utils_taxonomy_WFO import validate_taxonomy_WFO
from vouchervision.utils_geolocate_HERE import validate_coordinates_here
from vouchervision.tool_wikipedia import WikipediaLinks
class OpenAIHandler:
RETRY_DELAY = 10 # Wait 10 seconds before retrying
MAX_RETRIES = 3 # Maximum number of retries
STARTING_TEMP = 0.5
TOKENIZER_NAME = 'gpt-4'
VENDOR = 'openai'
def __init__(self, logger, model_name, JSON_dict_structure, is_azure, llm_object):
self.logger = logger
self.model_name = model_name
self.JSON_dict_structure = JSON_dict_structure
self.is_azure = is_azure
self.llm_object = llm_object
self.name_parts = self.model_name.split('-')
self.monitor = SystemLoadMonitor(logger)
self.has_GPU = torch.cuda.is_available()
self.starting_temp = float(self.STARTING_TEMP)
self.temp_increment = float(0.2)
self.adjust_temp = self.starting_temp
# Set up a parser
self.parser = JsonOutputParser()
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):
self.config = {'max_new_tokens': 1024,
'temperature': self.starting_temp,
'random_seed': 2023,
'top_p': 1,
}
# Adjusting the LLM settings based on whether Azure is used
if self.is_azure:
self.llm_object.deployment_name = self.model_name
self.llm_object.model_name = self.model_name
else:
self.llm_object = None
self._build_model_chain_parser()
# Define a function to format the input for azure_call
def format_input_for_azure(self, prompt_text):
msg = HumanMessage(content=prompt_text.text)
# self.llm_object.temperature = self.config.get('temperature')
return self.llm_object(messages=[msg])
def _adjust_config(self):
new_temp = self.adjust_temp + self.temp_increment
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):
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):
if not self.is_azure and ('instruct' in self.name_parts):
# Set up the retry parser with 3 retries
self.retry_parser = RetryWithErrorOutputParser.from_llm(
# parser=self.parser, llm=self.llm_object if self.is_azure else OpenAI(temperature=self.config.get('temperature'), model=self.model_name), max_retries=self.MAX_RETRIES
parser=self.parser, llm=self.llm_object if self.is_azure else OpenAI(model=self.model_name), max_retries=self.MAX_RETRIES
)
else:
# Set up the retry parser with 3 retries
self.retry_parser = RetryWithErrorOutputParser.from_llm(
# parser=self.parser, llm=self.llm_object if self.is_azure else ChatOpenAI(temperature=self.config.get('temperature'), model=self.model_name), max_retries=self.MAX_RETRIES
parser=self.parser, llm=self.llm_object if self.is_azure else ChatOpenAI(model=self.model_name), max_retries=self.MAX_RETRIES
)
# Prepare the chain
if not self.is_azure and ('instruct' in self.name_parts):
# self.chain = self.prompt | (self.format_input_for_azure if self.is_azure else OpenAI(temperature=self.config.get('temperature'), model=self.model_name))
self.chain = self.prompt | (self.format_input_for_azure if self.is_azure else OpenAI(model=self.model_name))
else:
# self.chain = self.prompt | (self.format_input_for_azure if self.is_azure else ChatOpenAI(temperature=self.config.get('temperature'), model=self.model_name))
self.chain = self.prompt | (self.format_input_for_azure if self.is_azure else ChatOpenAI(model=self.model_name))
def call_llm_api_OpenAI(self, prompt_template, json_report, paths):
_____, ____, _, __, ___, json_file_path_wiki, txt_file_path_ind_prompt = paths
self.json_report = 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})
response_text = response.content if not isinstance(response, str) else response
# Use retry_parser to parse the response with retry logic
output = self.retry_parser.parse_with_prompt(response_text, prompt_value=prompt_template)
if output is None:
self.logger.error(f'[Attempt {ind}] Failed to extract JSON from:\n{response_text}')
self._adjust_config()
else:
nt_in = count_tokens(prompt_template, self.VENDOR, self.TOKENIZER_NAME)
nt_out = count_tokens(response_text, 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_text}')
self._adjust_config()
else:
self.monitor.stop_inference_timer() # Starts tool timer too
json_report.set_text(text_main=f'Working on WFO, Geolocation, Links')
output, WFO_record = validate_taxonomy_WFO(output, replace_if_success_wfo=False) ###################################### make this configurable
output, GEO_record = validate_coordinates_here(output, replace_if_success_geo=False) ###################################### make this configurable
Wiki = WikipediaLinks(json_file_path_wiki)
Wiki.gather_wikipedia_results(output)
save_individual_prompt(Wiki.sanitize(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()
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")
self.json_report.set_text(text_main=f'Failed to extract valid JSON after [{ind}] attempts')
usage_report = self.monitor.stop_monitoring_report_usage()
self._reset_config()
json_report.set_text(text_main=f'LLM call failed')
return None, nt_in, nt_out, None, None, usage_report
|