VTimeLLM/vtimellm/utils.py

import datetime
import logging
import logging.handlers
import os
import sys
import time

import requests

from vtimellm.constants import LOGDIR

server_error_msg = "**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**"
moderation_msg = "YOUR INPUT VIOLATES OUR CONTENT MODERATION GUIDELINES. PLEASE TRY AGAIN."

handler = None


def build_logger(logger_name, logger_filename):
    global handler

    formatter = logging.Formatter(
        fmt="%(asctime)s | %(levelname)s | %(name)s | %(message)s",
        datefmt="%Y-%m-%d %H:%M:%S",
    )

    # Set the format of root handlers
    if not logging.getLogger().handlers:
        logging.basicConfig(level=logging.INFO)
    logging.getLogger().handlers[0].setFormatter(formatter)

    # Redirect stdout and stderr to loggers
    stdout_logger = logging.getLogger("stdout")
    stdout_logger.setLevel(logging.INFO)
    sl = StreamToLogger(stdout_logger, logging.INFO)
    sys.stdout = sl

    stderr_logger = logging.getLogger("stderr")
    stderr_logger.setLevel(logging.ERROR)
    sl = StreamToLogger(stderr_logger, logging.ERROR)
    sys.stderr = sl

    # Get logger
    logger = logging.getLogger(logger_name)
    logger.setLevel(logging.INFO)

    # Add a file handler for all loggers
    if handler is None:
        os.makedirs(LOGDIR, exist_ok=True)
        filename = os.path.join(LOGDIR, logger_filename)
        handler = logging.handlers.TimedRotatingFileHandler(
            filename, when='D', utc=True)
        handler.setFormatter(formatter)

        for name, item in logging.root.manager.loggerDict.items():
            if isinstance(item, logging.Logger):
                item.addHandler(handler)

    return logger


class StreamToLogger(object):
    """
    Fake file-like stream object that redirects writes to a logger instance.
    """
    def __init__(self, logger, log_level=logging.INFO):
        self.terminal = sys.stdout
        self.logger = logger
        self.log_level = log_level
        self.linebuf = ''

    def __getattr__(self, attr):
        return getattr(self.terminal, attr)

    def write(self, buf):
        temp_linebuf = self.linebuf + buf
        self.linebuf = ''
        for line in temp_linebuf.splitlines(True):
            # From the io.TextIOWrapper docs:
            #   On output, if newline is None, any '\n' characters written
            #   are translated to the system default line separator.
            # By default sys.stdout.write() expects '\n' newlines and then
            # translates them so this is still cross platform.
            if line[-1] == '\n':
                self.logger.log(self.log_level, line.rstrip())
            else:
                self.linebuf += line

    def flush(self):
        if self.linebuf != '':
            self.logger.log(self.log_level, self.linebuf.rstrip())
        self.linebuf = ''


def disable_torch_init():
    """
    Disable the redundant torch default initialization to accelerate model creation.
    """
    import torch
    setattr(torch.nn.Linear, "reset_parameters", lambda self: None)
    setattr(torch.nn.LayerNorm, "reset_parameters", lambda self: None)


def violates_moderation(text):
    """
    Check whether the text violates OpenAI moderation API.
    """
    url = "https://api.openai.com/v1/moderations"
    headers = {"Content-Type": "application/json",
               "Authorization": "Bearer " + os.environ["OPENAI_API_KEY"]}
    text = text.replace("\n", "")
    data = "{" + '"input": ' + f'"{text}"' + "}"
    data = data.encode("utf-8")
    try:
        ret = requests.post(url, headers=headers, data=data, timeout=5)
        flagged = ret.json()["results"][0]["flagged"]
    except requests.exceptions.RequestException as e:
        flagged = False
    except KeyError as e:
        flagged = False

    return flagged


def pretty_print_semaphore(semaphore):
    if semaphore is None:
        return "None"
    return f"Semaphore(value={semaphore._value}, locked={semaphore.locked()})"

def get_gpu_status():
    """
    Check the gpu stats and return the valid number of available gpus
    """
    from gpustat.core import GPUStatCollection

    gpus_stats = GPUStatCollection.new_query()

    info = gpus_stats.jsonify()["gpus"]
    gpu_list = []

    mem_ratio_threshold = 0.1  #
    util_ratio_threshold = 10  #
    for idx, each in enumerate(info):
        mem_ratio = each["memory.used"] / each["memory.total"]
        util_ratio = each["utilization.gpu"]
        if mem_ratio < mem_ratio_threshold and util_ratio < util_ratio_threshold:
            gpu_list.append(idx)
    print("Scan GPUs to get {} free GPU ({})".format(len(gpu_list), gpu_list))
    return gpu_list

def check_gpu_status(gpu_option):
    if gpu_option == 'cuda':
        env_var = 'CUDA_VISIBLE_DEVICES'
    elif gpu_option == 'gpu_vis':
        env_var = 'gpu_vis'
    print(f'gpu option is {env_var}')
    gpu_list = [int(x) for x in os.environ[env_var].split(',')]
    if len(gpu_list) == 0:
        assert False, 'Please specify the gpu_vis in the environment variable with export.'
    available_gpu = get_gpu_status()
    while gpu_list != available_gpu:
        print("No available GPU, waiting for 1 minutes until {} get freed. Current time : {}".format(gpu_list, time.ctime()))
        time.sleep(60)
        available_gpu = get_gpu_status()
    print("GPU is available now. Current time : {}".format(time.ctime()))