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