0.1.5

define python<3.13 since torch does not support 3.13

v0.1.4

Changed type hints to be compatible with Python 3.9.

v0.1.3

Added project source to the package.

v0.1.2

Decreased the batch size so that subwiz uses even less memory (1GB) with the same time performance.

v0.1.1

subwiz was crashing when using a large number of prediction due to out-of-memory. we now do inference in batches of 500 sequences at a time. in the future we could:

• make batch size configurable
• make batch size set automatically from memory size
• use quantization to decrease memory usage

v0.1.0

Publish