PyTorch 原型示例¶

Created On: Mar 03, 2021 | Last Updated: May 15, 2025 | Last Verified: Nov 05, 2024

原型功能在二进制分发（例如 PyPI 或 Conda）中不可用（可能除非启用运行时标志）。为了测试这些功能，根据不同的功能，我们建议从主版本构建或使用在 pytorch.org 上提供的 nightly wheels。

承诺等级: 我们承诺只对这些功能收集高带宽反馈。基于此反馈以及社区成员的进一步参与，我们作为一个社区将决定是否升级承诺等级或快速失败。

FX Graph Mode Quantization User Guide

Learn about FX Graph Mode Quantization.

FX,Quantization

FX Graph Mode Post Training Dynamic Quantization

Learn how to do post training dynamic quantization in graph mode based on torch.fx.

FX,Quantization

FX Graph Mode Post Training Static Quantization

Learn how to do post training static quantization in graph mode based on torch.fx.

FX,Quantization

Graph Mode Dynamic Quantization on BERT

Learn how to do post training dynamic quantization with graph mode quantization on BERT models.

Text,Quantization

PyTorch Numeric Suite Tutorial

Learn how to use the PyTorch Numeric Suite to support quantization debugging efforts.

Debugging,Quantization

How to Write a Quantizer for PyTorch 2 Export Quantization

Learn how to implement a Quantizer for PT2 Export Quantization

Quantization

PyTorch 2 Export Post Training Quantization

Learn how to use Post Training Quantization in PyTorch 2 Export.

Quantization

PyTorch 2 Export Quantization-Aware Training

Learn how to use Quantization-Aware-Training in PyTorch 2 Export.

Quantization

PyTorch 2 Export Quantization with X86 Backend through Inductor

Learn how to use PT2 Export Quantization with X86 Backend through Inductor.

Quantization

PyTorch 2 Export Quantization for OpenVINO torch.compile Backend

Learn how to use PT2 Export Quantization with OpenVINO torch.compile Backend.

Quantization

PyTorch 2 Export Quantization with Intel GPU Backend through Inductor

Learn how to use PT2 Export Quantization with Intel GPU Backend through Inductor.

Quantization

(prototype) Accelerating BERT with semi-structured (2:4) sparsity

Prune BERT to be 2:4 sparse and accelerate for inference.

Model-Optimiziation

Use iOS GPU in PyTorch

Learn how to run your models on iOS GPU.

Mobile

Convert MobileNetV2 to NNAPI

Learn how to prepare a computer vision model to use Android’s Neural Networks API (NNAPI).

Mobile

PyTorch Vulkan Backend User Workflow

Learn how to use the Vulkan backend on mobile GPUs.

Mobile

Tracing-based Selective Build Mobile Interpreter in Android and iOS

Learn how to optimize the mobile interpreter size with a tracing-based selective build.

Mobile

Convert Mobilenetv2 to Core ML

Learn how to prepare a computer vision model to use the PyTorch Core ML mobile backend.

Mobile

Skipping Module Parameter Initialization in PyTorch 1.10

Describes skipping parameter initialization during module construction in PyTorch 1.10, avoiding wasted computation.

Modules

Model Freezing in TorchScript

Freezing is the process of inlining Pytorch module parameters and attributes values into the TorchScript internal representation.

TorchScript

Using torch.vmap

Learn about torch.vmap, an autovectorizer for PyTorch operations.

vmap

Nested Tensor

Learn about nested tensors, the new way to batch heterogeneous-length data

NestedTensor

MaskedTensor Overview

Learn about masked tensors, the source of truth for specified and unspecified values

MaskedTensor

Masked Tensor Sparsity

Learn about how to leverage sparse layouts (e.g. COO and CSR) in MaskedTensor

MaskedTensor

Masked Tensor Advanced Semantics

Learn more about Masked Tensor's advanced semantics (reductions and comparing vs. NumPy's MaskedArray)

MaskedTensor

MaskedTensor: Simplifying Adagrad Sparse Semantics

See a showcase on how masked tensors can enable sparse semantics and provide for a cleaner dev experience

MaskedTensor

Inductor Cpp Wrapper Tutorial

Speed up your models with Inductor Cpp Wrapper

Model-Optimization

Inductor Windows CPU Tutorial

Speed up your models with Inductor On Windows CPU

Model-Optimization

Use max-autotune compilation on CPU to gain additional performance boost

Tutorial for max-autotune mode on CPU to gain additional performance boost

Model-Optimization

Flight Recorder Tutorial

Debug stuck jobs easily with Flight Recorder

Distributed, Debugging, FlightRecorder

Context Parallel Tutorial

Parallelize the attention computation along sequence dimension

Distributed, Context Parallel

Out-of-tree extension autoloading in Python

Learn how to improve the seamless integration of out-of-tree extension with PyTorch based on the autoloading mechanism.

Extending-PyTorch, Frontend-APIs

(prototype) Using GPUDirect Storage

Learn how to use GPUDirect Storage in PyTorch.

GPUDirect-Storage

PyTorch 原型示例¶

FX Graph Mode Quantization User Guide

FX Graph Mode Post Training Dynamic Quantization

FX Graph Mode Post Training Static Quantization

Graph Mode Dynamic Quantization on BERT

PyTorch Numeric Suite Tutorial

How to Write a Quantizer for PyTorch 2 Export Quantization

PyTorch 2 Export Post Training Quantization

PyTorch 2 Export Quantization-Aware Training

PyTorch 2 Export Quantization with X86 Backend through Inductor

PyTorch 2 Export Quantization for OpenVINO torch.compile Backend

PyTorch 2 Export Quantization with Intel GPU Backend through Inductor

(prototype) Accelerating BERT with semi-structured (2:4) sparsity

Use iOS GPU in PyTorch

Convert MobileNetV2 to NNAPI

PyTorch Vulkan Backend User Workflow

Tracing-based Selective Build Mobile Interpreter in Android and iOS

Convert Mobilenetv2 to Core ML

Skipping Module Parameter Initialization in PyTorch 1.10

Model Freezing in TorchScript

Using torch.vmap

Nested Tensor

MaskedTensor Overview

Masked Tensor Sparsity

Masked Tensor Advanced Semantics

MaskedTensor: Simplifying Adagrad Sparse Semantics

Inductor Cpp Wrapper Tutorial

Inductor Windows CPU Tutorial

Use max-autotune compilation on CPU to gain additional performance boost

Flight Recorder Tutorial

Context Parallel Tutorial

Out-of-tree extension autoloading in Python

(prototype) Using GPUDirect Storage

文档

教程

资源