VectorAlign Release 0.2.2

What's New

Memory Optimization

• Streaming processing: Processes sentences in batches instead of loading all embeddings into memory
• Should handle 1M+ sentences without freezing.
• Automatic garbage collection and CUDA cache clearing after each batch

Output Improvements

• Dictionary files now include header row (src tgt freq)

Bug Fixes

• Fixed [_save_dictionary] crash when output path has no directory
• Fixed duplicate tqdm progress bars

Installation

pip install vectoralign==0.2.2