Spaces:
Running
Running
File size: 6,684 Bytes
e91ac58 |
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 |
import os, time, random, torch
from langchain_mistralai.chat_models import ChatMistralAI
from langchain.output_parsers import RetryWithErrorOutputParser
from langchain.prompts import PromptTemplate
from langchain_core.output_parsers import JsonOutputParser
from vouchervision.utils_LLM import SystemLoadMonitor, count_tokens
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
class MistralHandler:
RETRY_DELAY = 2 # Wait 10 seconds before retrying
MAX_RETRIES = 5 # Maximum number of retries
STARTING_TEMP = 0.1
TOKENIZER_NAME = None
VENDOR = 'mistral'
RANDOM_SEED = 2023
def __init__(self, logger, model_name, JSON_dict_structure):
self.logger = logger
self.monitor = SystemLoadMonitor(logger)
self.has_GPU = torch.cuda.is_available()
self.model_name = model_name
self.JSON_dict_structure = JSON_dict_structure
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()
# 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):
self.config = {'max_tokens': 1024,
'temperature': self.starting_temp,
'random_seed': self.RANDOM_SEED,
'safe_mode': False,
'top_p': 1,
}
self._build_model_chain_parser()
def _adjust_config(self):
new_temp = self.adjust_temp + self.temp_increment
self.config['random_seed'] = random.randint(1, 1000)
self.json_report.set_text(text_main=f'Incrementing temperature from {self.adjust_temp} to {new_temp} and random_seed to {self.config.get("random_seed")}')
self.logger.info(f'Incrementing temperature from {self.adjust_temp} to {new_temp} and random_seed to {self.config.get("random_seed")}')
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} and random_seed to {self.RANDOM_SEED}')
self.logger.info(f'Incrementing temperature from {self.adjust_temp} to {self.starting_temp} and random_seed to {self.RANDOM_SEED}')
self.adjust_temp = self.starting_temp
self.config['temperature'] = self.starting_temp
self.config['random_seed'] = self.RANDOM_SEED
def _build_model_chain_parser(self):
# Initialize MistralAI
self.llm_model = ChatMistralAI(mistral_api_key=os.environ.get("MISTRAL_API_KEY"),
model=self.model_name,
max_tokens=self.config.get('max_tokens'),
safe_mode=self.config.get('safe_mode'),
top_p=self.config.get('top_p'))
# Set up the retry parser with the runnable
self.retry_parser = RetryWithErrorOutputParser.from_llm(parser=self.parser, llm=self.llm_model, max_retries=self.MAX_RETRIES)
self.chain = self.prompt | self.llm_model
def call_llm_api_MistralAI(self, prompt_template, json_report):
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, "random_seed": self.config.get("random_seed")}
# Invoke the chain to generate prompt text
response = self.chain.invoke({"query": prompt_template, "model_kwargs": model_kwargs})
# Use retry_parser to parse the response with retry logic
output = self.retry_parser.parse_with_prompt(response.content, prompt_value=prompt_template)
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.content, 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:
json_report.set_text(text_main=f'Working on WFO and Geolocation')
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
self.logger.info(f"Formatted JSON: {output}")
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
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')
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
|