batchtensor

0.1.1 verified Fri May 01 auth: no python

A Python library providing utility functions to manipulate batches of PyTorch tensors, such as concatenation, indexing, and nested data operations. Current version 0.1.1 supports Python 3.10-3.14 and PyTorch. The library follows a semantic versioning release cadence with regular updates.

pip install batchtensor

Common errors

error ImportError: cannot import name 'nested_apply' from 'batchtensor' ↓

cause Incorrect import path: nested_apply is in batchtensor.nested subpackage.

fix

Use: from batchtensor.nested import nested_apply

error AttributeError: module 'batchtensor' has no attribute 'cat_along_batch' ↓

cause Attempting to import after a partial installation or using an older version that did not export cat_along_batch at top level.

fix

Install latest version: pip install --upgrade batchtensor

Warnings

breaking Version 0.0.4 introduced breaking changes (see release notes). Functions may have been renamed or arguments changed. ↓

fix Upgrade and update function calls per changelog.

gotcha batchtensor uses nested functions (e.g., nested_apply, nested_map) that expect nested structures of tensors. Passing plain tensors may raise unexpected errors. ↓

fix Ensure input is a dict or list of tensors, not a single tensor.

deprecated Some functions may be deprecated in future releases. Check the release notes before upgrading. ↓

fix Review deprecation warnings and migrate to recommended alternatives.

Imports

cat_along_batch
```
from batchtensor import cat_along_batch
```
Common mistake: trying to import from submodules like 'batchtensor.nested' before checking top-level exports.
nested_apply
wrong
```
from batchtensor import nested_apply
```
correct
```
from batchtensor.nested import nested_apply
```
nested_apply is in the nested submodule, not directly in batchtensor.

Quickstart

Basic usage: concatenate a list of tensors along the batch dimension.

import torch
from batchtensor import cat_along_batch

tensor1 = torch.randn(2, 3)
tensor2 = torch.randn(3, 3)
result = cat_along_batch([tensor1, tensor2])
print(result.shape)  # torch.Size([5, 3])