v0.3.0

What's Changed

Faster model training

We made model training faster by applying JAX's JIT compilation, pooling file writes, etc.

• Faster training data encoding by @tushuhei in #89
• Add out_span option for better GPU utilization by @tushuhei in #90
• Apply JAX JIT compiling for faster training by @tushuhei in #95
• Check in updated Simplified Chinese model by @tushuhei in #99

Smaller models

We made models smaller by removing less important features, disabling ASCII encoding, etc.

• Remove Unicode Block features by @tushuhei in #86
• Disable ASCII encoding when building the model file by @tushuhei in #98
• Output compact model by @tushuhei in #100

Misc

• encode_data: write without break line join by @tushuhei in #91
• Update unit tests for the encoding script by @tushuhei in #92
• Add more granularity in weight outputs by @tushuhei in #93
• Remove tar module dependency by @tushuhei in #96

Full Changelog: v0.2.1...v0.3.0