wandb_bug_reproduce.py

wandb_bug_reproduce.py

import wandb
import torch.multiprocessing as mp

wandb.init(project="test", entity="leoleoasd", group="123")

def train(group):
    wandb.init(project="test", entity="leoleoasd", group=group)
    wandb.log({
        'loss': 1
    })
    wandb.finish()

procs = []
barrier = mp.Barrier(8)
for i in range(8):
    proc = mp.Process(target=train, args=("123",))
    procs.append(proc)
    proc.start()
for proc in procs:
    proc.join()