WandB Run Identifier (naive, untested example) to exemplify the idea

wandb_run_identifier.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import time, logging, platform
import wandb

def wandb_init(opt, *args, **kwargs):
    """A quick naive example """
    wandb_identifier = get_run_identifier(opt, *args, **kwargs)
    wandb.init(project="aProj", name=wandb_identifier, notes="<Intriguing notes>")
    wandb.config.update(opt)
    model = kwargs.get('model')
    if model:
        wandb.watch(model)
    # Init a Table of Stats|Results
    stats = kwargs.get('stats')
    if stats:
        stats_table = wandb.Table(columns=list(stats.keys()))


def get_run_identifier(opt, *args, **kwargs) -> str:
    """
    A unique run identifier for runs and corresponding checkpoints with the same time-stamp
    Usage: @see wandb_init
    Use as (prefix|suffix) E.g. f"{project_name}|{identifier}"

    :param args: Run arguments
    :return: A identifier string for a run set
    """
    ts = time.strftime('%Y%m%d-%H%M%S')
    template = kwargs.get('template', 'default_template')
    if opt.is_train:
        train_phase = "reinforce" if opt.is_finetune else "pretrain"
        identifier = f"[ {template} ]-reason.run_train-{train_phase}-{opt.num_iters}num_iters.{opt.max_train_samples}tr_samples" \
                     f".{opt.max_val_samples}val_samples.{opt.batch_size}bsz-{platform.node()}-{ts}"
    else:
        guess = "-GUESS" if opt.exec_guess_ans else ""
        identifier = f"[ {template} ]-reason.run_test{guess}-[{opt.load_checkpoint_path}]" \
                     f".{opt.batch_size}bsz-{platform.node()}-{ts}"
    logging.debug("-" * 100)
    logging.debug(f"Run identifier: {identifier}")
    logging.debug("-" * 100)

    return identifier