Lovely Tensors

0.1.22 · active · verified Fri Apr 17

lovely-tensors is a Python library that provides "syntax sugar" for PyTorch tensors, enhancing their default `repr` for better readability, debugging, and development. It adds features like rich, colored, and pretty-printed output, automatic batching, and shape inference. As of version 0.1.22, it is actively maintained with frequent minor releases focusing on new features and improvements.

Common errors

```
ModuleNotFoundError: No module named 'lovely_tensors'
```
cause The `lovely-tensors` library is not installed in your current Python environment.

fix Run `pip install lovely-tensors` to install the library.
```
PyTorch tensors are not displaying with pretty colors or enhanced information, even after importing lovely_tensors.
```
cause The `monkey_patch()` function was not called, which is necessary to enable the global `__repr__` enhancement.

fix After `import lovely_tensors as lt`, add `lt.monkey_patch()` to activate the desired behavior.
```
AttributeError: 'Tensor' object has no attribute 'lovely'
```
cause You are trying to access `lovely` as a method directly on a `torch.Tensor` instance, but `lovely` is a function provided by the `lovely_tensors` module, not a tensor method.

fix Instead of `my_tensor.lovely()`, use `lt.lovely(my_tensor)` to explicitly pretty-print the tensor.

Warnings

gotcha To activate the enhanced `__repr__` for PyTorch tensors automatically, you must explicitly call `lovely_tensors.monkey_patch()` after importing the library. Simply importing `lovely_tensors` is not enough.
Fix: Ensure you have `import lovely_tensors as lt` followed by `lt.monkey_patch()` in your initialization code, typically at the start of your script or notebook.
gotcha lovely-tensors modifies the `__repr__` method of `torch.Tensor` instances. If you are using other debugging tools or libraries that also monkey-patch `torch.Tensor`'s `__repr__`, this might lead to conflicts or unexpected display behavior.
Fix: If conflicts arise, try to import `lovely-tensors` before other potentially conflicting libraries. Alternatively, avoid `lt.monkey_patch()` and use `lt.lovely(tensor_instance)` for explicit pretty-printing when needed, giving you more control.
deprecated Older versions of lovely-tensors used `lovely_tensors.init()` for activation. This method has been replaced by `lovely_tensors.monkey_patch()` for clarity and consistency.
Fix: Update your code to use `lt.monkey_patch()` instead of `lt.init()` for global activation.

Install

pip install lovely-tensors Install latest version

Imports

lovely_tensors
```
import lovely_tensors as lt
```
The common alias `lt` is widely used in examples and makes the code more concise.
monkey_patch
wrong:
```
import lovely_tensors.monkey_patch # This doesn't activate it
```
correct:
```
import lovely_tensors as lt
lt.monkey_patch()
```
Calling `lt.monkey_patch()` directly is necessary to enable the enhanced `__repr__` for PyTorch tensors globally.
lovely
```
import lovely_tensors as lt
lt.lovely(my_tensor)
```
Use `lt.lovely()` for explicit, controlled pretty-printing of a tensor, or when monkey-patching is not desired.

Quickstart

This quickstart demonstrates how to import `lovely-tensors`, enable its global monkey-patching for `torch.Tensor`'s `__repr__`, and then print PyTorch tensors to see the enhanced, colored output. It also shows how to use `lt.lovely()` for explicit pretty-printing.

import torch
import lovely_tensors as lt

# Enable beautiful tensor repr globally
lt.monkey_patch()

# Create some tensors
a = torch.randn(2, 3, 4)
b = torch.randint(0, 10, (1, 5), dtype=torch.int)

print(f"Tensor a:\n{a}")
print(f"Tensor b:\n{b}")

# You can also explicitly print lovely tensors without monkey-patching
# For demonstration, let's pretend monkey_patch wasn't called:
# import lovely_tensors as lt
# lt.lovely(torch.tensor([1,2,3]))

view raw JSON →