---
description: Optimize your PyTorch models with Ultralytics YOLO's torch_utils functions such as ModelEMA, select_device, and is_parallel.
---

# ModelEMA
---
:::ultralytics.yolo.utils.torch_utils.ModelEMA
<br><br>

# EarlyStopping
---
:::ultralytics.yolo.utils.torch_utils.EarlyStopping
<br><br>

# torch_distributed_zero_first
---
:::ultralytics.yolo.utils.torch_utils.torch_distributed_zero_first
<br><br>

# smart_inference_mode
---
:::ultralytics.yolo.utils.torch_utils.smart_inference_mode
<br><br>

# select_device
---
:::ultralytics.yolo.utils.torch_utils.select_device
<br><br>

# time_sync
---
:::ultralytics.yolo.utils.torch_utils.time_sync
<br><br>

# fuse_conv_and_bn
---
:::ultralytics.yolo.utils.torch_utils.fuse_conv_and_bn
<br><br>

# fuse_deconv_and_bn
---
:::ultralytics.yolo.utils.torch_utils.fuse_deconv_and_bn
<br><br>

# model_info
---
:::ultralytics.yolo.utils.torch_utils.model_info
<br><br>

# get_num_params
---
:::ultralytics.yolo.utils.torch_utils.get_num_params
<br><br>

# get_num_gradients
---
:::ultralytics.yolo.utils.torch_utils.get_num_gradients
<br><br>

# get_flops
---
:::ultralytics.yolo.utils.torch_utils.get_flops
<br><br>

# initialize_weights
---
:::ultralytics.yolo.utils.torch_utils.initialize_weights
<br><br>

# scale_img
---
:::ultralytics.yolo.utils.torch_utils.scale_img
<br><br>

# make_divisible
---
:::ultralytics.yolo.utils.torch_utils.make_divisible
<br><br>

# copy_attr
---
:::ultralytics.yolo.utils.torch_utils.copy_attr
<br><br>

# get_latest_opset
---
:::ultralytics.yolo.utils.torch_utils.get_latest_opset
<br><br>

# intersect_dicts
---
:::ultralytics.yolo.utils.torch_utils.intersect_dicts
<br><br>

# is_parallel
---
:::ultralytics.yolo.utils.torch_utils.is_parallel
<br><br>

# de_parallel
---
:::ultralytics.yolo.utils.torch_utils.de_parallel
<br><br>

# one_cycle
---
:::ultralytics.yolo.utils.torch_utils.one_cycle
<br><br>

# init_seeds
---
:::ultralytics.yolo.utils.torch_utils.init_seeds
<br><br>

# strip_optimizer
---
:::ultralytics.yolo.utils.torch_utils.strip_optimizer
<br><br>

# profile
---
:::ultralytics.yolo.utils.torch_utils.profile
<br><br>