This article is a mirror article of machine translation, please click here to jump to the original article.

Architect_Programmer_Code Agriculture Network»Architect › Programming › Technical chat › Common models for deep learning (.pt, . onnx) file format ...
View: 1922|Reply: 1

Common models for deep learning (.pt, . onnx) file format

[Copy link]
Posted on 2025-2-4 20:43:40 | | |
PyTorch
File format: .pt or .pth
Description: Used to hold the weights, schema, and other metadata of the PyTorch model. Usually used in the training and reasoning phases.

TorchScript
File format: .pt or .torchscript
Description: A PyTorch model compiled with TorchScript that can run without a Python interpreter. Ideal for efficient model deployment and cross-platform applications.

ONNX Runtime
File format: .onnx
Description: ONNX (Open Neural Network Exchange) is an open format used to represent deep learning models. ONNX Runtime is a high-performance inference engine that supports a variety of hardware accelerators. ONNX files can be exported by multiple frameworks and used on multiple platforms.

OpenVINO
File formats:.xml and .bin
Description: The OpenVINO toolkit converts a model into an intermediate representation (IR) with two files:
.xml File: Describes the structure of the model.
.bin File: Stores the weights of the model. OpenVINO supports multiple hardware accelerations such as CPU, GPU, VPU, etc.

TensorRT
File format: .engine
Description: TensorRT is a high-performance inference optimization library provided by NVIDIA that is specifically optimized for GPUs. engine files are TensorRT-optimized models for efficient inference on NVIDIA GPUs.

CoreML (macOS-only)
File format: .mlmodel
Description: CoreML is a machine learning framework provided by Apple designed for iOS and macOS devices. mlmodel files are models in CoreML format that allow for efficient inference directly on Apple devices.

TensorFlow GraphDef
File format: .pb
Description: GraphDef is an early model saving format for TensorFlow, which primarily contains definitions of computational graphs. Although not as popular as SavedModel, it is still used in some cases.

TensorFlow Lite
File format: .tflite
Description: TensorFlow Lite is a lightweight version of TensorFlow designed for mobile and embedded devices. TFlite files are optimized models for running in resource-constrained environments.

PaddlePaddle
File formats: .pdparams and .pdmodel
Description: PaddlePaddle is an open-source deep learning framework developed by Baidu. Models are usually saved as two files:
.pdmodel file: describes the architecture of the model.
.pdparams file: Stores the weights of the model.




Previous:"Kokoro-82M" is a text-to-speech TTS model that has recently exploded
Next:[AI] (2) The difference between DeepSeek-V3 vs R1 versions

Related Posts
 Landlord| Posted on 2025-2-7 10:52:01 |
[AI] (6) A brief introduction to the large model file format GGUF
https://www.itsvse.com/thread-10936-1-1.html
Disclaimer:
All software, programming materials or articles published by Code Farmer Network are only for learning and research purposes; The above content shall not be used for commercial or illegal purposes, otherwise, users shall bear all consequences. The information on this site comes from the Internet, and copyright disputes have nothing to do with this site. You must completely delete the above content from your computer within 24 hours of downloading. If you like the program, please support genuine software, purchase registration, and get better genuine services. If there is any infringement, please contact us by email.
Mail To:help@itsvse.com