HeyTensor

What is HeyTensor?

HeyTensor is a free, browser-based tool designed to help developers quickly calculate PyTorch tensor shapes without executing code. It offers 22 specialized tools that can determine the output dimensions for various PyTorch layers, including common operations like Convolution, Linear, LSTM, and Transformer layers. This allows developers to visualize and verify tensor transformations at different stages of their neural network architecture.

The calculator includes pre-defined architecture presets, such as LeNet, a ResNet block, and a Transformer encoder, allowing users to quickly test common network structures. A unique "paste error" mode is also available, which parses PyTorch RuntimeError messages to help identify shape-related issues. This feature is particularly useful for debugging unexpected tensor dimension mismatches during model development and training.

This tool is ideal for machine learning engineers and researchers working with PyTorch who need to understand how data flows through their models. By providing instant shape calculations, HeyTensor aims to reduce the time spent on manual shape verification and debugging, enabling faster iteration on model designs. Its accessibility as a free, browser-based application makes it a convenient resource for quick checks and educational purposes.

Key features

• 22 tensor calculation tools.
• Supports Conv, Linear, LSTM, Transformer.
• Architecture presets included.
• Paste error mode.
• Browser-based interface.
• No code execution needed.
• Free to use.
• PyTorch focused.

Use cases

• Verifying tensor shapes before writing model code.
• Debugging shape mismatch errors in PyTorch models.
• Understanding tensor transformations in neural network layers.
• Quickly checking output dimensions for common architectures.
• Educating oneself on PyTorch tensor flow.

Pros & cons

FAQ