---
description: Explore valuable torch utilities from Ultralytics for optimized model performance, including device selection, model fusion, and inference optimization.
keywords: Ultralytics, torch utils, model optimization, device selection, inference optimization, model fusion, CPU info, PyTorch tools
---

# Reference for `ultralytics/utils/torch_utils.py`

!!! success "Improvements"

    This page is sourced from [https://github.com/ultralytics/ultralytics/blob/main/ultralytics/utils/torch_utils.py](https://github.com/ultralytics/ultralytics/blob/main/ultralytics/utils/torch_utils.py). Have an improvement or example to add? Open a [Pull Request](https://docs.ultralytics.com/help/contributing/) — thank you! 🙏

<br>

## ::: ultralytics.utils.torch_utils.ModelEMA

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.EarlyStopping

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.torch_distributed_zero_first

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.smart_inference_mode

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.autocast

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.get_cpu_info

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.get_gpu_info

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.select_device

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.time_sync

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.fuse_conv_and_bn

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.fuse_deconv_and_bn

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.model_info

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.get_num_params

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.get_num_gradients

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.model_info_for_loggers

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.get_flops

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.get_flops_with_torch_profiler

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.initialize_weights

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.scale_img

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.copy_attr

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.intersect_dicts

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.is_parallel

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.unwrap_model

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.one_cycle

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.init_seeds

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.unset_deterministic

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.strip_optimizer

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.convert_optimizer_state_dict_to_fp16

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.cuda_memory_usage

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.profile_ops

<br><br><hr><br>

## ::: ultralytics.utils.torch_utils.attempt_compile

<br><br>