Flops-Counter.Pytorch

Visit Tool

flops-counter.pytorch is an Open Source & Models tool that computes theoretical multiply-add operations and parameters for neural networks in PyTorch. It also provides per-layer computational cost analysis.

Claim this tool

1View

At a glance

Pricing

Open Source

Free tier

Yes

API

Skill level

Technical

About

What is flops-counter.pytorch?

flops-counter.pytorch is an open-source tool designed to calculate the theoretical number of multiply-add operations (FLOPs) and parameters within neural networks built using the PyTorch framework. It offers two backends: 'pytorch' for legacy nn.Modules with better per-layer analytics for CNNs, and 'aten' for broader coverage of model architectures, including transformers, by considering aten operations. The tool can also print per-layer computational costs and allows for ignoring specific modules during counting. It supports various layers like Conv1d/2d/3d, BatchNorm, Activations, Linear, Upsample, and Poolings, with experimental support for RNNs, LSTMs, GRUs, and MultiheadAttention. Users can customize input tensors for complex models and view verbose output for unconsidered operations.

Best used for

Ideal for developers and data scientists who need to analyze the computational complexity of their PyTorch neural networks, benchmark different model architectures, and optimize network efficiency. Especially valuable for researchers and practitioners working with deep learning models to understand their resource requirements.

Common actions

analyze model complexity

benchmark neural networks

optimize PyTorch models

count FLOPs

evaluate model efficiency

workflowsopen-sourcecollaborationdeepfakegithub copilotlow-code/no-code"AI Agents"automated workflowface swapping

Capabilities

Key features

FLOPs counting
Parameter counting
Per-layer cost analysis
PyTorch backend
Aten backend
Ignore specific modules
Custom input tensors

Target Audience

developerdata scientist

Integrations

Not yet documented

Pricing & Plans

Open Source

Free

FAQs

What is the difference between the 'pytorch' and 'aten' backends?

The 'pytorch' backend is a legacy option that primarily considers nn.Modules and offers better per-layer analytics for CNNs. The 'aten' backend is recommended for most cases, as it considers aten operations, covering a wider range of model architectures, including transformers.

Can I use flops-counter.pytorch with transformer models?

Yes, for transformer architectures, it is strongly recommended to use the 'aten' backend. The 'pytorch' backend is not suitable for transformers as it may not accurately account for all operations in these complex models.

How can I ignore specific modules from the FLOPs count?

You can use the `ignore_modules` option in the `get_model_complexity_info` function. This allows you to specify a list of modules, such as `torch.nn.Conv2d` or `torch.ops.aten.convolution`, that should be excluded from the computational cost calculation.

Trending

Subcategories trending in Coding & Development

Code Assistants DevOps & Infrastructure No-Code / Low-Code Testing & QA Backend & APIs Prompt Engineering

Trending

Explore

Browse AI tools by category

Content & Design Productivity & Business Coding & Development AI Agents & Automation Research & Education Wellness & Lifestyle Career Development Marketing & Growth Data & Analytics Customer Support & CX Finance E-commerce