Why does the logging config in model.py not take effect?

[DongZhaoXiong]

Description

I define a logger class like this:

import logging
import json
import os, sys
from datetime import datetime
class LLMLogger:
    def __init__(self, name="LLMApp", log_level=logging.INFO, json_format=False):
        self.logger = logging.getLogger(name)
        self.logger.setLevel(log_level)
        self.json_format = json_format

        log_format = "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
        formatter = logging.Formatter(log_format)
        
        console_handler = logging.StreamHandler()
        console_handler.setFormatter(formatter)
        self.logger.addHandler(console_handler)

    def _format_message(self, message, **kwargs):
        if self.json_format:
            log_entry = {
                "timestamp": datetime.utcnow().isoformat(),
                "message": message,
                "context": kwargs
            }
            return json.dumps(log_entry)
        return f"{message} | Context: {kwargs}"

    def debug(self, message, **kwargs):
        self.logger.debug(self._format_message(message, **kwargs))

    def info(self, message, **kwargs):
        self.logger.info(self._format_message(message, **kwargs))

    def warning(self, message, **kwargs):
        self.logger.warning(self._format_message(message, **kwargs))

    def error(self, message, **kwargs):
        self.logger.error(self._format_message(message, **kwargs))

    def critical(self, message, **kwargs):
        self.logger.critical(self._format_message(message, **kwargs))
if __name__ == "__main__":
    logger = LLMLogger(name="LLMProductionApp", log_level=logging.DEBUG, json_format=True)

    logger.info("LLM service launch", service="LLMModel", version="v1.0")
    logger.warning("high workload", queue_size=50, requests_in_queue=20)
    logger.error("generate failed", request_id="12345", error="Timeout")
    logger.debug("debug message", computation_time=0.123)

Expected Behavior

python3 local/logger.py
2024-12-24 07:52:19,956 - LLMProductionApp - INFO - {"timestamp": "2024-12-24T07:52:19.956471", "message": "LLM service launch", "context": {"service": "LLMModel", "version": "v1.0"}}
2024-12-24 07:52:19,956 - LLMProductionApp - WARNING - {"timestamp": "2024-12-24T07:52:19.956633", "message": "high workload", "context": {"queue_size": 50, "requests_in_queue": 20}}
2024-12-24 07:52:19,956 - LLMProductionApp - ERROR - {"timestamp": "2024-12-24T07:52:19.956727", "message": "generate failed", "context": {"request_id": "12345", "error": "Timeout"}}
2024-12-24 07:52:19,956 - LLMProductionApp - DEBUG - {"timestamp": "2024-12-24T07:52:19.956808", "message": "debug message", "context": {"computation_time": 0.123}}

Error Message

WARN PyProcess W-190-model-stderr: 2024-12-24 07:53:52,585 - LLMProductionApp - INFO - {"timestamp": "2024-12-24T07:53:52.585513", "message": "============================load model successfully====================", "context": {}}
INFO PyProcess W-190-model-stdout: INFO::{"timestamp": "2024-12-24T07:53:52.585513", "message": "============================load model successfully====================", "context": {}}

How to Reproduce?

custom_logger = LLMLogger(name="LLMProductionApp", log_level=logging.DEBUG, json_format=True)

def load_model(properties):
    custom_logger.info(f"Model configuration:{properties}")

    if "model_id" in properties:
        model_location = properties['model_id']

    custom_logger.info(f"Loading model in {model_location}")
    
    tokenizer = AutoTokenizer.from_pretrained(model_location, trust_remote_code=True)

    custom_logger.info("============================load tokenizer successfully====================")                                    
    
    model = AutoModelForCausalLM.from_pretrained(
        model_location, 
        device_map="auto",
        trust_remote_code=True,
        torch_dtype="auto"
    )

    custom_logger.info('============================load model successfully====================')     

    return model, tokenizer

why has duplicate output logging?