could stop training

[laohutigerzhang]

Search before asking

• I have searched the RF-DETR issues and found no similar bug report.

Bug

when I train this rf-detr model, it cannot stop training because it trained 100/100 epoch it have not stopped but continue training from 1/100 epoch …… I don't know why!

Environment

rfdetr==1.5.1
windows10
python==3.14.3
pytorch==2.10.0+cu130
CUDA/cuDNN == 13.0
GPU RTX 3060 laptop
CPU  AMD Ryzen 7 5800H

Minimal Reproducible Example

The training code is as follows:

import requests
import supervision as sv
from PIL import Image
from rfdetr import RFDETRSegMedium
import torch

model.train(
dataset_dir=r"XXX",
epochs=100,
batch_size=2,
grad_accum_steps=4,
lr=1e-4,
output_dir="./results",
device="cuda",
early_stopping=True
)

Additional

No response

Are you willing to submit a PR?

• Yes, I'd like to help by submitting a PR!

[dain20000222]

I also got the similar issue. Could you try changing num_workers = 0 ?

See the bellow issue:
#688

[Vishalkagade]

On windows, if num_workers > 0, the spawned worker re-execuates the script. To bypass it put that into main guard.

if __name__ == '__main__':
    model = RFDETRSegMedium()
    model.train(
        dataset_dir=r"XXX",
        epochs=100,
        batch_size=2,
        grad_accum_steps=4,
        lr=1e-4,
        output_dir="./results",
        device="cuda",
        early_stopping=True,
    )